The reason why this particular feature (click on image) doesn't download the file is because that's what a significant amount of the users requested. It's a poor man's zoom, so to speak, and it's reproducing a workflow that people used to have on the file page and missed in media viewer.
I think we're paying the price of implementing this hack as a placeholder for zoom. To me the point where we have to build convoluted mechanisms (especially extra menus, warnings...) to compensate for a strange feature is usually a sign that it shouldn't be built like this in the first place.
Pau proposed a good compromise. I.e. clicking should open a bigger image (the "zoomed" version) but not the original, unless the original happens to be smaller than what we consider to be the bigger version. I think that's a smarter placeholder hack than what we're currently doing. It's worth a shot.