Ben Bitdiddle gets really excited about remote procedure call and implements an RPC package himself. His implementation starts a new service thread for each arriving request. The thread performs the operation specified in the request, sends a reply, and terminates. After measuring the RPC performance, Ben decides that it needs some improvement, so Ben comes up with a brute-force solution: he buys a much faster network. The transit time of the new network is half as large as it was before. Ben measures the performance of small RPCs (meaning that each RPC message contains only a few bytes of data) on the new network. To his surprise, the performance is barely improved. What might be the reason that his RPCs are not twice as fast?.