I can see what you are saying. I would say however that most $$ players in the stock market have analytic teams, etc. If instead, a substantial portion of the players in the stock market placed their bets on randomly chosen stocks, the aggregate would be much less efficient and that could be exploited.

I've sat and watched people betting on horses for a long time. The majority choose based on the name or color of the horse (sentimentalist), the going favorite (risk averse), the longest odd (big paydayists). Many others play on weaker signals (owner, jockey). A few bet on the advice of experts in the daily form, and these probably do make the market more efficient. In aggregate, from my own experience at Golden Gate Fields where the take-out is 14%, my average ROI was around -6 to -10%. This suggests that I was beating the market, but the takeout was killing me.

So I've imagined that in a decentralized paramutual pool with minuscule or zero takeout, and given a common population of betters, I'd make a steady profit.

I should add that another advantage of paramutual over bookmaker odds is the pool maintainers do not care if you are a winner or a loser, and won't freeze your accounts on you.