I have only programmed access to db engines on occasion.
But someone asked a question that sounded like a "worst case
scenario" for a db engine. Namely the person has a word list,
like a spelling dictionary. Due to lack of maintanence it
has grown large with many duplicates. Since usually one thinks
of looking up a chunk of data by the key, it would seem that a
"keys-only database" would be slow and unwieldy. Have you run
across a similar application? The dictionary is a flat text
file with a word per line. The size has grown many times that
of physical ram on the machine. Therefore attempting to sort
as a means to remove duplicates would take days if not weeks of
merge sort.

Do you have any suggestions or approaches? I assume the person is running Windows XP. I suggested a Windows port of the Linux sort commands. Started off fine but after 6 or 7 days and still churning the user gave up and killed it.