Bitorrent kind of does this anyway, because in order to download one file, you need to download the entire block it's in. In a case like this where you have a large torrent with many small files, that will actually cover quite a few other files. If you ordered them strategically, you could even try to make sure every block has a fairly popular file in it.
Yeah, good idea. Sort the papers by popularity, and when grouping them find the distribution that minimizes the distance between any paper and the top-popular ones for atomic block size.
I don't think you need to do anything that complex. Just find the N most popular papers and make sure there's one in each block.
Or thinking about, you could put them at block boundaries so people grab two blocks when they download them, in which case you'd only need half as many popular ones.