diff --git a/xtask/src/main.rs b/xtask/src/main.rs
index ee399901..db6cbabe 100644
--- a/xtask/src/main.rs
+++ b/xtask/src/main.rs
@@ -567,6 +567,14 @@ impl Step {
     const BENCH_VALIDATE: Self = Self {
         name: "bench (validate)",
         run: || {
+            // Use the dev profile (not the default bench profile) for the
+            // criterion `--test` smoke run. The bench profile inherits the
+            // release profile's `lto = true, codegen-units = 1`, which spends
+            // ~40s compiling the full graph just to assert criterion's
+            // `--test` entry point runs once. The dev profile reuses the
+            // already-built unit-test artifacts and finishes in seconds. Real
+            // benchmark measurements (via `cargo xtask bench`) continue to
+            // use the unchanged bench profile so their numbers stay accurate.
             run_command_quiet(
                 "cargo",
                 &[
@@ -575,6 +583,7 @@ impl Step {
                     "microsoft-webui",
                     "--bench",
                     "contact_book_bench",
+                    "--profile=dev",
                     "--",
                     "--test",
                 ],
diff --git a/xtask/src/util.rs b/xtask/src/util.rs
index 931fc592..6b636b0b 100644
--- a/xtask/src/util.rs
+++ b/xtask/src/util.rs
@@ -72,7 +72,25 @@ pub fn run_command_quiet(cmd: &str, args: &[&str], cwd: Option<&Path>) -> Result
 /// On Windows `CreateProcessW` cannot launch `.cmd`/`.bat` scripts directly.
 /// This function uses `which` to resolve the executable path and, when the
 /// target is a shell script, wraps it in `cmd.exe /c <resolved_path>`.
+///
+/// For cargo invocations, this also forces `CARGO_INCREMENTAL=0`. xtask is
+/// only used for one-shot work (gate, examples, bench, packaging, E2E) and
+/// never for an inner edit-compile loop, so incremental compilation provides
+/// no benefit. Leaving it on causes the workspace `target/` dir to balloon
+/// (the incremental cache and per-feature fingerprint variants grow into the
+/// tens of gigabytes), which then makes every subsequent cargo invocation pay
+/// huge startup time scanning hundreds of thousands of stale fingerprint
+/// files. Disabling it here keeps `cargo xtask check` reliably fast on
+/// long-lived working copies.
 pub fn build_command(cmd: &str, args: &[&str]) -> Command {
+    let mut command = build_command_inner(cmd, args);
+    if cmd == "cargo" {
+        command.env("CARGO_INCREMENTAL", "0");
+    }
+    command
+}
+
+fn build_command_inner(cmd: &str, args: &[&str]) -> Command {
     #[cfg(windows)]
     {
         resolve_windows_command(cmd, args)