I suspect if data from eye-tracking tests were available, there would be an extremely clear revealed preference from users. I read image galleries the exact same way I skim text, in an ordered fashion that allows me to "read" every image without reading an image twice, stopping if my attention is caught by something in particular. Splotting garbage over the screen haphazardly makes it blend together annoyingly and results in my eyes traversing the same areas multiple times both to try to pick out details and to try to keep my place in what I have/haven't skimmed yet. It is a layout that itself demands my attention, rather than letting my attention be absorbed naturally by the actual images.
From actual eye tracking data via Hotjar and similar, people do skip around the page. Those that scan linearly are in the minority but probably are more highly represented on HN, just as a matter of course.