Scanned page images *shouldn't* be the best way to read books (unless they contain a lot of layout). If PG (and e-reader programs, for that matter) handled images decently they would not be. But they are.
One can find plenty of awful page image books too -- page images containing images which require a great deal of digital image manipulation to get them to a "readable" [or viewable] state. PG offers images in three sizes right now: 1) zero sized -- no images at all. 2) what the volunteer HTML submitter chose based on their best understanding of what would be best for the PG community -- or perhaps based on what they ate for breakfast that morning. [I know I am reluctant to submit books over 10 meg in size -- due to downloading speed/cost issues for e-book readers on slow cellphone connections.] 3) what the PG html to epub/mobi converter software does to that html file. One could image a downloading system which resizes images "on the fly" based on the target machine. Amazon already does that -- not sure their system has been very successful. But this is starting to sound like the forever-ongoing "big machine / small machine" arguments....