Expanding somewhat on the problem of not using real money - if people lose all their money they can't bet any more. This combats obstinance. If people are simply too stubborn to update on evidence then they will no longer be able to bet. Their poor judgement is removed from the market. The only way this would come in to play on lesswrong is if the bets started to become seriously large. If the system was karma transfer (which I recommended against) then poorly calibrated predictors would lose their karma reserves. This would benefit the quality of large stake predictions but it would completely alter the meaning implied by karma.

Errr... now that I read that back to myself it tempts me towards perverse speculation that that the change in implicit meaning of karma could be a good thing. At least to the extent that karma can be used as a metric for quality of expected predictions (only a minor purpose).