We can fairly easily add the ability to track latency, IO, etc... Between RPC clients, which can be incredibly useful if you're in a situation where you want to see what client is slowing down an operation and vice-versa. Also, who doesn't like numbers?