diff --git a/CHANGELOG.md b/CHANGELOG.md
index a392162f..a2d42ef6 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -15,6 +15,8 @@
 - `--jobs auto` / `-j auto` caps parallel concurrency at the CPU core count (portable across Linux/macOS/BSD); the default stays unlimited (#766)
 
 ### Changed
+- Faster test runs: stripping ANSI codes from short colored strings is pure bash now, so aligning the per-test execution time (shown on systems with a fork-free clock, e.g. Linux) no longer forks `sed` once per passing test
+- Faster parallel runs: publishing each test's result file no longer forks `basename`, `mkdir` and an `echo | tr | sed` pipeline per test — the suite dir name is parameter expansion, the dir is pre-created once per file before workers spawn, and arg sanitizing is skipped without provider args (a 10-test parallel run dropped from 61 to 21 forks). No behaviour change
 - Faster test runs: listing all defined functions uses the `compgen -A function` builtin instead of forking `declare -F | awk` (three call sites; 5 -> 3 `awk` forks per test file). No behaviour change
 - Faster test runs: ordering a file's test functions by definition line is pure bash now instead of an `awk | sort | awk` pipeline that ran twice per file, and the duplicate-function check sorts its (usually empty) result inside awk instead of piping through `sort` (9 -> 5 forks per test file). No behaviour change
 - Faster failure rendering: the "Source:" assert-line context of a failing test reads the test-function body in a single pass instead of forking `sed` per line and `grep` per line to find the closing brace (a 20-line body dropped from ~46 to ~2 forks here). No behaviour change
diff --git a/src/math.sh b/src/math.sh
index b0b69462..cc3df1eb 100644
--- a/src/math.sh
+++ b/src/math.sh
@@ -19,8 +19,14 @@ function bashunit::math::calculate() {
     ;;
   esac
 
-  # Remove leading zeros from integers
-  expr=$(echo "$expr" | sed -E 's/\b0*([1-9][0-9]*)/\1/g')
+  # Remove leading zeros from integers so $((...)) does not read them as octal.
+  # Only fork sed when a leading zero is actually present — the common callers
+  # (clock durations) never produce one, so the no-bc path stays fork-free.
+  case "$expr" in
+  0[0-9]* | *[!0-9.]0[0-9]*)
+    expr=$(echo "$expr" | sed -E 's/\b0*([1-9][0-9]*)/\1/g')
+    ;;
+  esac
 
   local result=$((expr))
   echo "$result"
diff --git a/src/runner.sh b/src/runner.sh
index e0572d7a..1a2d914d 100755
--- a/src/runner.sh
+++ b/src/runner.sh
@@ -838,6 +838,14 @@ function bashunit::runner::call_test_functions() {
     allow_test_parallel=false
   fi
 
+  # Pre-create the file's result dir before spawning test workers: they all
+  # publish into it, and checking `[ -d ]` inside a worker races its siblings
+  # (every worker would still pay the mkdir fork).
+  if bashunit::parallel::is_enabled && [ "$allow_test_parallel" = true ]; then
+    local _suite_base="${script##*/}"
+    mkdir -p "${TEMP_DIR_PARALLEL_TEST_SUITE}/${_suite_base%.sh}" 2>/dev/null || true
+  fi
+
   for fn_name in "${functions_to_run[@]+"${functions_to_run[@]}"}"; do
     if bashunit::parallel::is_enabled && bashunit::parallel::must_stop_on_failure; then
       break
@@ -1560,11 +1568,19 @@ function bashunit::runner::parse_result_parallel() {
   local -a args
   args=("$@")
 
-  local test_suite_dir="${TEMP_DIR_PARALLEL_TEST_SUITE}/$(basename "$test_file" .sh)"
-  mkdir -p "$test_suite_dir"
+  # This runs once per test in every parallel worker, so avoid per-test forks:
+  # derive the suite dir name with parameter expansion (no basename), only
+  # mkdir when the dir is missing (first test of the file wins the race,
+  # `-p` makes the losers no-ops), and skip arg sanitizing entirely for the
+  # common no-provider-args case.
+  local test_suite_base="${test_file##*/}"
+  local test_suite_dir="${TEMP_DIR_PARALLEL_TEST_SUITE}/${test_suite_base%.sh}"
+  [ -d "$test_suite_dir" ] || mkdir -p "$test_suite_dir"
 
-  local sanitized_args
-  sanitized_args=$(echo "${args[*]+"${args[*]}"}" | tr '[:upper:]' '[:lower:]' | sed -E 's/[^a-z0-9]+/-/g; s/^-|-$//')
+  local sanitized_args=""
+  if [ -n "${args[*]+"${args[*]}"}" ]; then
+    sanitized_args=$(echo "${args[*]}" | tr '[:upper:]' '[:lower:]' | sed -E 's/[^a-z0-9]+/-/g; s/^-|-$//')
+  fi
   local template
   if [ -z "$sanitized_args" ]; then
     template="${fn_name}.XXXXXX"
diff --git a/src/str.sh b/src/str.sh
index 7b828264..90143eec 100644
--- a/src/str.sh
+++ b/src/str.sh
@@ -17,6 +17,50 @@ function bashunit::str::strip_ansi_to_slot() {
     return
     ;;
   esac
+  # Pure-bash path for short strings without backslashes: display lines (e.g.
+  # the per-test "✓ Passed … 3ms" alignment on systems whose clock is fork-free)
+  # land here, so a colored line does not cost a sed fork per test. Strip
+  # CSI sequences segment-wise, then sweep remaining control bytes; the size
+  # guard avoids bash's quadratic pattern-substitution on large captures and
+  # `*\\*` still defers to `echo -e` semantics below.
+  case "$input" in
+  *\\*) ;;
+  *)
+    if [ "${#input}" -le 1024 ]; then
+      local out="" rest="$input" params
+      while :; do
+        case "$rest" in
+        *$'\x1b'\[*)
+          out="$out${rest%%$'\x1b'\[*}"
+          rest="${rest#*$'\x1b'\[}"
+          params=""
+          while :; do
+            case "$rest" in
+            [0-9\;]*)
+              params="$params${rest%"${rest#?}"}"
+              rest="${rest#?}"
+              ;;
+            *) break ;;
+            esac
+          done
+          case "$rest" in
+          # Same finals sed strips; anything else keeps its printable residue
+          # (the ESC itself falls to the control-byte sweep, exactly like sed).
+          m* | K*) rest="${rest#?}" ;;
+          *) out="${out}[${params}" ;;
+          esac
+          ;;
+        *)
+          out="$out$rest"
+          break
+          ;;
+        esac
+      done
+      _BASHUNIT_STR_STRIPPED_OUT=${out//[[:cntrl:]]/}
+      return
+    fi
+    ;;
+  esac
   _BASHUNIT_STR_STRIPPED_OUT=$(echo -e "$input" | sed -E 's/\x1B\[[0-9;]*[mK]//g; s/[[:cntrl:]]//g')
 }
 
diff --git a/tests/acceptance/bashunit_run_forks_test.sh b/tests/acceptance/bashunit_run_forks_test.sh
index abb39224..f8703a56 100644
--- a/tests/acceptance/bashunit_run_forks_test.sh
+++ b/tests/acceptance/bashunit_run_forks_test.sh
@@ -182,3 +182,49 @@ function test_run_removes_its_run_output_dir() {
 
   assert_equals 0 "$leftover"
 }
+
+# Regression guard for the parallel per-test result path. Publishing each
+# test's result file used to fork `basename` (suite dir name), `mkdir -p`
+# (suite dir, per test) and an `echo | tr | sed` pipeline (arg sanitizing, even
+# with no args) — ~4 forks per test in the mode CI runs everything in. The dir
+# name is parameter expansion now, mkdir is guarded by a `[ -d ]` builtin check,
+# and arg sanitizing is skipped when there are no provider args.
+function test_parallel_result_publishing_does_not_fork_per_test() {
+  if bashunit::check_os::is_windows; then
+    bashunit::skip "PATH shims are unreliable under Git Bash" && return
+  fi
+
+  local dir
+  dir="$(bashunit::temp_dir)"
+  local count_file="$dir/count"
+  local bin
+  for bin in basename tr sed; do
+    local real_bin
+    real_bin="$(command -v "$bin")"
+    {
+      echo '#!/usr/bin/env bash'
+      echo "echo $bin >> \"$count_file\""
+      echo "exec \"$real_bin\" \"\$@\""
+    } >"$dir/$bin"
+    chmod +x "$dir/$bin"
+  done
+
+  local fixture="$dir/parallel_forks_test.sh"
+  {
+    echo 'function test_a() { assert_true true; }'
+    echo 'function test_b() { assert_true true; }'
+    echo 'function test_c() { assert_true true; }'
+    echo 'function test_d() { assert_true true; }'
+  } >"$fixture"
+
+  PATH="$dir:$PATH" ./bashunit --parallel "$fixture" >/dev/null 2>&1
+
+  # Assert on the shim log's content, not a count: a failure then names the
+  # offending binary directly in the test output.
+  local forked=""
+  if [ -f "$count_file" ]; then
+    forked="$(sort "$count_file" | uniq -c | tr -d '\n')"
+  fi
+
+  assert_equals "" "$forked"
+}
diff --git a/tests/unit/str_test.sh b/tests/unit/str_test.sh
index 8f52c52c..3e4547d8 100644
--- a/tests/unit/str_test.sh
+++ b/tests/unit/str_test.sh
@@ -128,3 +128,31 @@ function test_rpad_width_smaller_than_right_word() {
 
   assert_same "... verylongword" "$actual"
 }
+
+function test_strip_ansi_to_slot_removes_erase_codes_and_control_chars() {
+  local input
+  input="$(printf '\033[2K\033[1;32mok\033[0m\tdone\r')"
+
+  bashunit::str::strip_ansi_to_slot "$input"
+
+  assert_same "okdone" "$_BASHUNIT_STR_STRIPPED_OUT"
+}
+
+function test_strip_ansi_to_slot_long_input_matches_short_path() {
+  # Inputs beyond the pure-bash size guard take the sed path; both paths must
+  # produce identical output for the same (repeated) colored payload.
+  local unit="\033[31mred\033[0m plain "
+  local long_input=""
+  local short_expected=""
+  local n=100
+  while [ "$n" -gt 0 ]; do
+    long_input="${long_input}${unit}"
+    short_expected="${short_expected}red plain "
+    n=$((n - 1))
+  done
+  long_input="$(printf '%b' "$long_input")"
+
+  bashunit::str::strip_ansi_to_slot "$long_input"
+
+  assert_same "$short_expected" "$_BASHUNIT_STR_STRIPPED_OUT"
+}