The thread-per-request approach