— Small memory footprint even for large datasets.

— ACID transactions.

— SQL interface for introspection and reporting.

That's sometimes even just for development work.

A lot of these use a common API to a more complex distributed store, as well as to something simple like files on disk, in memory, or SQLite.

I'm most cases, it's one user at a time, so performance doesn't matter, but simplicity does.

It can also be for the project which has a 1 percent chance of going viral.

Etc. But I find relatively few cases between truly small scale and large scale.