We can fix the DBM problem by going to another database system and plugging that into a the libc shared library. This might take a four days of work by an experienced Unix hacker, but should otherwise be fairly easy to deploy. It would buy us about six months, at most, assuming the number of simultaneous users does not increase.
The absolute number of user ids could be fixed with a new database, and some other smarts: one suggestion was to build virtual machines with chroot(2), and loopback mounts, and isolate the different user ids.
Another suggestion, implemented in the CIX freeport software is to do away with unique user ids, and run all non-guest, non-shell users with the same user id. This could be fairly easy, particularly with some libc hacking on access(2).