In the words of one genius:
Databases aren't magical. Quite the reverse. They easily end up being _slower_ than doing it by hand, simply because they have to solve a much more generic issue. If you design your data structures and abstractions right, a database is pretty much guaranteed to only incur overhead. - Linus Torvalds
MPD is already really fast. It doesn't scale to 100,000 songs at the drop of a hat, but that's fine, and that will probably improve given a bit of time as well.
This has been one of the bigger issues of the last year. Every non-coder comes in and wants to use *SQL to store MPD's database. It's not that easy, as I'm sure a few have tried. None have been successful yet. If you think you can do it, branch/fork the code, make something usable, stable, as fast as or faster than the MPD database, and _then_ we'll talk about using an *SQL as an alternate database for MPD.