I think you're right, in a typical database situation.

But one could imagine a case where the database is small enough to fit in memory. If that is combined with a read-mostly database, the above could be a good fit.

I wish I had the time and the hardware to make a good benchmarked comparison, but alas.