Loading rolldown in a Node worker_thread, letting it exit, then loading rolldown in a second worker_thread in the same process segfaults the host (SIGSEGV / exit 139). A single worker is fine; the crash appears on the 2nd. It's deterministic in this minimal form.

npm install
node main.mjs; echo "exit=$?"   # → exit=139 (SIGSEGV); "PASS" never prints

• worker.mjs — loads rolldown, runs a trivial in-memory build, closes, exits.
• main.mjs — spawns the worker twice, sequentially, awaiting each one's natural exit.

• rolldown 1.1.0 (latest; what this repro pins) and 1.0.3 — both crash
• Node 20, 22, 24 — all crash
• macOS, darwin-arm64

• One worker never crashes; the crash needs a second rolldown load in the same process after the first worker has exited.
• Replacing the worker_thread with a child_process (separate process) does not crash.
• The same harness with @swc/core (also Rust + napi + native threads) instead of rolldown does not crash — so it's not a generic "native addon in a worker_thread" problem; it's specific to rolldown.

strings on the native binary (@rolldown/binding-darwin-arm64) contains thread local panicked on drop, aborting, Rayon: detected unexpected panic; aborting, and the symbols start_async_runtime / shutdown_async_runtime — which points at native runtime teardown across worker_thread lifecycles. No symbolicated backtrace captured, so the root-cause is left to the maintainers.