I had precisely the same reaction about the persistent effects of torture over time when I read the torture vs. dust specks problem.

I think your reply to the original question highlights a difficulty with actual application of utilitarian thought experiments in concrete situations. The question as originally posed involved inflicting disutility on a random target by pushing a button, presumably with the actor and the target being mutually unaware of each other's identities. When you substitute punching someone, even if the target is randomly chosen, the thought experiment breaks down, both because it becomes difficult to predict the actual amount of suffering that is going to result (e.g. the target has a black belt and/or concealed weapon and no sense of humor = more pain than you were expecting), and because of second order effects: a world in which inflicting random pain is considered justified if it produces a greater amount of fun for other individuals is going to be a world with a lot less social trust, reducing utility for everybody.