For those following my plight with interest or amusement, I'm back with results of the switch to MLDBM::Sync...and those results are, disappointingly, even worse than the DB_File results.

I tried both explicit and implicit locking (explicit should be faster per the docs). So I'm locking and unlocking for each 256 objects entered into the DB. The difference is small but measurable. Anyway, I suspect my objects cross the line into >1024 byte territory too frequently to gain the benefit of this module over DB_File. I figured them for being smaller than that, but I must be wrong...I think I'll do an object size histogram of the data set for my own enlightenment for the next installment.

So, next verse, quite similar to the first...I'll be converting everything over to the newer BerkeleyDB module, with further experiments in locking and not locking. More profiling to come. And some day, a solution.