A lot of web requests are tens of millliseconds due to the latency of speaking to the database. The abolity to fire several requests to the database instead of serialising them is one of this optimisations that you really can’t do in threaded model without introducing other asynchronous systems components.