Have you measured how it performs, compared to a simple boost.asio based "thread-pool". Boost.asio is often used in the core of proprietary thread-pool implementations, so I am curious ;)