I was thinking about using something like DBM::Deep to avoid DB I/O, [sic] have you used it / [sic] do you know if it's any good?

Yes, I’ve used it. I wanted to see whether I could use it to replace the innards of something that was using MLDBM in conjunction with DB_File (or maybe it was BerkeleyDB?) and Storable underneath. I wanted to do this for the improved UI experience, since even with MLDBM you have to manage some of the multilevel stuff on your own, nontransparently.

It was all a terrible mistake: DBM::Deep ran so very very slowly, and was so huge, as to immediately disqualify itself from the running. (Plus it installed about 50 new modules, and not without problems either, on my system through CPAN recursion. Simply incredible the bloat people take for granted these days!) I don’t have the numbers on me right now, but it was a good bit more than even a full order of magnitude slower than doing it the other way. The performance just didn’t cut the mustard.

I was ready for a drop in performance, but I could not handle rather more than 1000% slower. And bigger. That just won’t work. If you don’t mind 10, 20, maybe 30 times slower, well, go for it. But don’t say I didn’t tell you so. Better to spend the time investigating the improved facilities of latterday BerkeleyDB releases instead.