The most complicated category would be moderately sized, low margin transactions, in a time sensitive physical retail environment. The archetypical example is a grocery store. With transactions possibly exceeding $300 the fraud risk becomes material. 0-confirm probably can't be relied upon.
Thank you for the thoughtful comments, DeathAndTaxes.
I agree with most of the points you made, but could you clarify why you think $300 grocery purchases may be too high a risk for zero-confirm transactions? Assuming that malleability is eliminated in the future, how could a customer using a mobile app cheat the grocery store cashier [assuming the grocery store has a well-connected listening node]? My understanding is that the fraudster would either need to coordinate a Sybil attack, or he would need to pass the fraudulent double-spend over a non-public back-channel to an unethical miner that controls a great deal of global hash power.
Both these attack methods seem like a lot of work for groceries, neither method is guaranteed to succeed, and the fraudster risks getting caught and charged with a crime. (Despite what Augusto Croppo says, as bitcoin becomes more mainstream, we will begin to recognize these shenanigans as a crime similar to using counterfeit bills or stolen credit cards).
I imagine that in a "bitcoin future" there will be less zero-confirm fraud using bitcoin [malleability aside] than purchases with counterfeit cash or stolen credit cards. Do you disagree?