In the current implementation each connection runs in its own goroutine. All the requests send using that connection get handled sequentially by that same routine. It might be possible to handle each request in its own goroutine and thus speeding up the server if resources are available.