If I were running a service like Carbonite, I would detect when my users were backing up the same file and not store multiple copies of the same thing. I’m sure people back up their music collections, and there’s probably a great deal of redundancy there. Files for which, based on the filename alone, it’s easy to find candidates for an exact match. A byte-by-byte comparison could determine whether two files are identical. Then? Just store one copy and have users’ backup sets which have that file just point to it. Essentially they’d be creating a pool of all songs downloadable or being traded online, and for each user store a reference to their subset of the pool.