The issue isn't that all miners would be fraudulent but rather a small % would.
The earlier post mentioned 20%. That to me seemed like a large percentage. Where-as if just 1% of miners are fraudulent, then the chances of a successful double-spend become quite low, compared to the meagre profit from ripping off the kind of merchants that would accept zero-confirmation transactions, and the cost of setting up the mining farm, and the damage to reputation when you get caught. I think we can get to a situation where it's irrational to try. I appreciate that doesn't mean no-one will, but it will be extremely rare.
I also appreciate you are saying that some merchants have such low tolerance for the risk that even "extremely rare" won't be good enough for them.
Some half-baked thoughts on this topic:
Very roughly, the % lost due to fraudulent double spends is proportional to the percentage of hash rate controlled by a fraudulent mining group willing to accept out-of-band transaction AND the percentage of the population that will download an app (and thereby knowingly engage in fraud) to transmit the out-of-band double spend. Mathematically:
% loss on 0-confirms = (% hash power controlled by fraud miners) x (% of consumers engaged in fraud).
If we assume 20% of the hash power is fraudulent (much more than I would expect) and that 10% of consumers will attempt to cheat every time (much more than I would expect), then the % loss on 0-confirms is 2%:
% loss on 0-confirms = (0.2) x (0.1) = 0.02 = 2%.
Like DeathAndTaxes said, 2% is too much for a grocery store, but not too much for Naughty America. However, I think more reasonable numbers would be that only 5% (or less) of the hash rate is fraudulent, and that perhaps only 2% of consumers (or less) would attempt to cheat. With these numbers we get only 0.1% losses (or less):
% loss on 0-confirms < (0.05) x (0.02) = 0.001 = 0.1%.
Some other interesting things I noted:
1. If the % of hash power accepting out-of-band double spends is 25%, double-spenders would receive one quarter of their purchases for free [of course, they still risk getting caught and charged with a crime, similar to using a stolen credit card or counterfeit bills]. However, as the % of hash power controlled by the "fraud miners" decreases, this service becomes a lot less tempting. If they only control 1% of hash rate, then double-spenders only succeed 1% of the time, in which case I think very few would even bother and the problem mostly disappears.
2. Another note is that it probably won't be a single group accepting the out-of-band double spends. A good double spend app, however, would want to transmit the fraudulent transaction to all the knowingly-complicit hash power, in order to maximize the odds of a successful double spend. Now, these fraud miners need to stay on the down-low, which means they will be moving around lots, etc., and the apps will need to be constantly updated with the latest info. But this confusion means that honest people have an opportunity to pose as fraud miners, just so that they can broadcast (i.e., leak) the double spends onto the network and neutralize the double-spend threat. In other words, the double-spend threat can be neutralized by leaking the fraudulent transaction.
It will be interesting to see how it all develops....