End-to-end (glass-to-glass) latency is substantially better. Mostly because the protocol isn't request/response any more.