I mentioned before that, in the days of 50 btc per block rewards, I solved one block, but did 3 blocks worth of work. On most pools I solved no blocks at all. Did I cheat those pools out of $50 000? Are you also 95% certain that the 5% of pool members that have had similarly bad luck are also cheating this pool?
Its clear to me this problem cant be solved with statistics. Variance of good faith small hashrates is far too great, and anyone with bad faith and large hashrate can easily divide his hashpower over as many accounts/workers as he wants.
I dont know how stratum works at all, but the only solution I can think off that would be 100% watertight is changing the protocol in such a way that the miner can not know if his share solves a block or a not, and only the pool can determine that. Is that something thats even theoretically possible to implement?