The key is potential. We can't have the computers hung up on every single node and it's possibilities, but what one system of 10 computers discards as possibly insignificant, the next system of 10 computers could find different results.
In the end, it's all about branching out like quantum computers, but we've got to figure out ways to minimize unnecessary lookups.
Here's a setup you could build with your friends, maybe it'll answer your question:
Connect 6 computers to a network (or build a blade server, etc.), have computer 1 & 2 calculate two separate branches of nodes and load them into a buffer, have computers 3, 4 & 5 running different engines working on those nodes to produce their own analysis, have computer 6 fine comb the results over a cup of coffee and decide what move is to be played. Stream the entire thing to us and post the link here.