The advantage of the web is that you are browsing the documents themselves rather than som structure hiding them. Since these documents link to each other you do not have to remember folds structures but simply click.
As Ajax becomes more prevalent, some sites and their features/documents are only accessible by some click sequence, rather than via direct navigation.
Well-behaved sites tend to be structured better, and offer navigational clues (breadcrumbing), and the like.
Many sites really don't have much intrinsic structure and are little more than a large bag (or set of bags) of pages or other objects (Facebook, YouTube, Wikipedia), though these may have linkages (perferably hypertextual) between objects (notably: Wikipedia and kin).
One of the things I'm liking about using an e-book reader (Moon+Reader on Android, Calibre on Debian/Linux) is that rather than scatter docs through the filesystem, they're organized through the reader interface itself. Not an entirely novel concept -- the man and info commands have used a similar concept for years, and are amenable to organization through tools such as Debian's dwww (online localhost Web interface to locally stored documentation, including manpages, info pages, package documentation, and installed documentation resources). Including search and indexing.
Where the "document" dynamically changes depending on actions th euser takes within it, which aren't revealed/exposed through the URL, then this falls apart.
