Even absent the cluster of machines I found this to be very useful. Several years ago, I would use zeromq (thank you, btw) for communication and if the socket overhead was found to be an issue, I would merge two processes and switch from tcp:// to inproc://. Many "corrupt the world" type bugs are isolated from each other by processes, and it made it very hard to accidentally share state (the file system of course is a giant hunk of share state, but you very rarely accidentally write to a file).