The dbd::csv and dbd::xbase modules will let you run your own mini database without installing any additional rdbms software. They will, however, create multiple files per db.

Is the single file requirement for runtime or for convenience? There are ways you could make it feel like a single file db... You could store your db files in a single tar file and then extract them at runtime, stuff them back in on program exit. This would give your code a very crude rollback functionality also.

What about running against dbd::ram and then using something like dbd::anydata to dump the entire contents to a file periodically?

Since you're going to take a performance hit with *any* sort of perl based db solution, you could just alter your design. Combine your tables into one big table and just query it for what you're looking for.