I am not convinced that if a fat man were actually standing there waiting to be shoved piteously onto the tracks that the human mind would necessarily function in the same way it does when sitting in a cafe and discussing the fate of said to-be switch-pusher.

If I were to stake the distinction between the actual and the theoretical on anything, it would be on the above point. What data have we on the reliability of these -- I think you must agree that, regardless of the hypothetical opinions of medieval scholar types, the Torture vs. Dust Specks scenario abstract for us now and here -- thought experiments to predict human behavior when, to retreat to the cliche, one is actually in the trenches?

I... don't think the point of such thought experiments was ever to predict what a human will do. That we do not make the same choices under pressure that we do when given reflection and distance is quite obvious. If you are interested in predicting what people will do, you should look at psychological battery tests, which (should) strive to strike a balance between realism and measurability.

The point of train tracks-type experiments was to force one to demand some coherence from their "moral intuition", and to this end the fact that you're making such choices sitting in a café is a feature, not a bug, because it lets you carefully figure out logical conclusions on which (at least in theory) you will then be able to unthinkingly rely once you're in the heat of the moment (probably not an actual train track scenario, but situations like giving money to beggars or voting during jury duty where you only have seconds or hours to make a choice). When you're actually in the trenches, as you put it, your brain is going to be overwhelmed by a zillion more cognitive biases than usual, so it's very much in your interest to try and pre-make as many choices as possible while you have the luxury of double-checking every one of your assumptions and implications.