> They're forking a new process for each request. Hm, I don't think they are. I think they one persistent process managing the state of each document, and many

> They're forking a new process for each request. Hm, I don't think they are. I think they one persistent process managing the state of each document, and many requests might interact with that document, via stdin/stdout channels on that document's process. Makes the case for the rest of what you said all the stronger, though.