Hacker News Reader

Either a) those pdfs contain images or other graphics that don't compress easily or b) the text is not compressed.

Or c) The text is terribly compressed. Probably a lot of the "text" is also a scan of text as well.

Could be an interesting project to try and make a sci-hub optimized compression system.

DejaVu format

There are definitely many images and graphics. I don't know if my sample is representative of most literature, but I don't think it's too far off..