The problem we had with that is the users who are most inclined to purchase ad-free memberships are also the users who generate dramatically more ad revenue than other users.
When you average it all out, you're counting users as if they all generate the same number of pageviews or clicks, but obviously that's not really the case. Your most active users are probably generating at least an order of magnitude more views and clicks than non-active users, but those same active users are the ones who buy memberships.