A better solution is not to use cost-per-click in the first place, instead relying on metrics that are harder to game, like unique visitors and page views. These are arguably a better reflection of an advertising space's value even without taking click fraud into account. That's what Project Wonderful did, and it has worked out well for some of its users, but its model is also more long-tail/long-tail, which I think hurts its profitability.
Surely page views are as easy to imitate as clicks (possibly easier). Unique visitors would be reasonably easy to game as well I'd imagine.
Say I'm a fraudster with a thousand IPs and a well-anonymized browser. If I'm paid per click, I can easily turn that into a thousand apparently legitimate clicks, which, depending on how I play my cards, could net me a few hundred dollars.
By contrast, if I'm paid per unique visitor, I can turn that into a thousand apparently legitimate unique visitors, which will very optimistically pull in maybe $5 (judging by Project Wonderful).
Now, neither of those schemes are optimal, but just about any improvement you suggest for the unique visitor model will reap ten times the rewards applied to the per-click model.