Your argument seems to me to conflate judgments that "X-ing is wrong" with predictions that one would not X if faced with a particular choice in real life.

Judgements that a choice is morally wrong are clearly not the same thing as predictions about whether people would make that choice. The way I view morality though a wide gulf between the two is indicative of a problem to be resolved. I see the purpose of morality as providing a framework for solving something analogous to an iterated prisoners dilemma. If we can all agree to impose certain restrictions on our own actions because we all expect to do better if everyone sticks to the rules then we have a system of morality.

Humans have a complex interplay of instinctive moral intuitions and cultural norms that together form a moral framework that exists because it provides a reasonably stable solution to living in mutually beneficial societies. That doesn't mean it can't be improved, just that its very existence implies that it works reasonably well.

The problem then with a moral dilemma that appears to present a wide gap between what people say should be done and what people would actually do is that it suggests a flaw in the moral framework. A stable framework will generally require that decisions that people can agree are right (in that we'd expect on average to be better off if we all followed them) are also decisions that people can plausibly commit to taking if faced with the problem. It's like the pre-commitment problem discussed before on less wrong. You might wish to argue for an idealized morality that sets standards for what people should do that are not what most people would do but then you have to make a plausible case for why what people actually do is wrong. Further, I'd argue you have to make a case for how your system could actually be implemented with actual people in a stable fashion - an idealized morality that is not achievable with actual people is not very interesting to me.

Ultimately I don't take a utilitarian view of morality - that what is 'good' is what maximizes utility across all agents. I take an 'enlightened self interest' view - that what is 'good' is what all agents can agree is a framework that will tend to lead to better expected outcomes for each individual if each individual constrains his own immediate self interest in certain ways.