Benchmarking aside, I'm supposing that you don't mind being forced into a programmatic mode of access for the files in question while using BerkleyDB? In other words, there's no reason to access the files in question with something as mundane as vi?

That's a huge consideration for me. As an example of how to go terribly wrong with this, consider what IBM did with AIX -- all system configuration files were mirrored in a database. As a consequence, most unix system configuration commands had an AIX-centric version that would ensure the database was synchronized with the flat files. Yuck.

Matt