Well, here's a plug for Durus.  I finally got my chemical database
program running (not the web interface, that's the next part). with
interchangeable database backends for Durus, SQLAlchemy (MySQL and
SQLite), and CSV files, plus older custom backends to MySQL and
SQLite.

With a 6100 record table with 200 fields, plus two search tables,
Durus is consistently 2-3 times faster than either MySQL or SQLite for
importing the records, scanning through all of them, or comparing one
database to another.    Plus there's no per-query overhead so you
don't have to choose whether to get the records one by one (slow in
SQL) or all at once (memory intensive) or in chunks of say 50
(kludgey) -- which means rewriting a queries for all backends when you
hit a SQL bottleneck.

I was quite amazed at this, but it shows up consistently through
several utility programs, backend implementations, and core classes
rewrite.  I haven't tried it with a high concurrency load yet, but if
it can handle that, I'm tempted to switch our other web applications
to it as they get revised.

PS. The reason for the multiple backends is so we can compare the
database technologies and switch at a later date if we want to.  The
same application will run on a webserver and on disconnected laptops.
The client wants MySQL on the server, but that has licensing and
installation issues on the laptops so we've been evaluating SQLite and
Durus as alternatives.  Plus I used Durus as the reference
implementation because it was so easy to write.

--
Mike Orr