Thanks - my bad. I confused your point about the challenge (or benefits) of integrating p2pool into the reference client ("…these cheap machines would still be out of the question… The main client really struggles on the 512Mb RAM raspberry pi") with your separate/main point regarding variance. Thanks for being patient with me as I become more familiar with the details of p2pool and the mining landscape.

It sounds like there are some good discussions happening among people who know a lot more about this than I, so I'm happy about that. I'm just hoping that an effective solution can be engineered, deployed, and adopted before it's too late