I usually do a Subject->Category->Author (or Sub-category)->Title thing for my 300GB+ collection. For instance, Novels/SciFi/Vance/BlueWorld.pdf, Math/Geometry/Differential/Foundations-Michor.djvu or Comics/Manga/Berserk/Chapter01.cbr. That way, novels, which you will probably want to access by Author, are indexed correctly but do not end up being too cluttered, as they are divided into SciFi, Fiction etc., while textbooks are still indexed by subject. Other unrelated things like comics and journal articles live in their own tree.
Unfortunately, with folder organization, you cannot cross list items (such as by both publisher and topic). I tried using shortcuts by keeping my *real* collection tucked away in a bunch of folders and indexing the collection in different ways with shortcuts, but creating it can be a real pain.
Otherwise, one would need to come up with an indexing application that lets you search/browse your library (and preferably integrate it into the CTB). I've never tried exploring it, but how powerful is FBReader's library? I don't think there are any universal "ID3 like" tags for e-books like there are for audio files, so we'd need something to index stuff intelligently.
For quick reference lists, I usually dump a bunch of shortcuts into a folder, like a playlist. This should keep you from traversing too many trees. The redundancy in the index hurts my sense of efficiency though, would love a better solution
Of course, once we get an SD card-wide search function onto the DR (should be zippy for pre-indexed content), we needn't bother with organizing stuff too much. Would love to see a 'recently used documents' list.