Hm. I hadn't even heard of Proof of Work filters. (Since Jon didn't provide an example link, I will: Kapow.) The basic idea is that the client is required to perform a complex programming task via JavaScript in the hopes that a real user is willing to wait a few seconds and a spammer is unlikely to want to spend CPU time to generate the appropriate response.

Personally, I question the validity of that argument. Undoubtedly, spammers currently don't have full JavaScript engines in their backends, so are currently failing those tests. But I doubt that that's going to be a huge concern if PoW filters become popular. I could be wrong.

I happen to think that ReCAPTCHA is a great solution. The PoW authors claim is that it's not difficult to programmatically decode the corrupted text. But they use text that has already failed to be decoded, because it's text that has failed OCR for a real text. This means that your effort is not going to wasted, as it's going to help OCR text in an automated turk fashion, and either the spambots are going to fail the test, or they're going to develop algorithms that can decode currently undecodable text. I think it's exceedingly clever social engineering.