That would mean ~2 cache lines are accessed for an IPC call, plus another 130 cycles for instructions. That's very impressive.