Well, I'd argue that any program for which the difference between a keys-based walk and an each-based walk is significant shouldn't be using hashes that large, not with the arrival of things like DBD::SQLite or other cheap databases.

Yes, it mattered when Perl was running on 10Mhz 1Megabyte machines, and databases were tough, so a DBM hash was a cool thing. Those are long gone. {grin}