I don't think that experiment makes sense: if there is a low quality copy for free and a higher quality copy for 5$ I take the lower quality copy to be for evaluation. So only 18% of the people who evaluated it decided to buy it. In fact even without the low/high quality thing, perhaps in a "pay what you want setting" people download first and evaluate. Perhaps those people who didn't pay simply didn't like the music.

Also, even in that experiment they seemed to have made almost 1$ per download - isn't that more than artists typically get per CD sale?