Is there a performance penalty using a pure Python database interface? Yes there is. But how much? .. also depends on who wrote it.
I started implementing some benchmarking for MySQL Connector/Python. Main reason is to identified bottlenecks or just plain bad coded. Another reasons: it's cool and the question was raised during my talk at FOSDEM and also online.
Oh, yes, MySQL Connector/Python ain't fast right now. Just compare a script spawning 10 threads opening 1000 connections. It indeed just connects.
mysql.connector 0.1.3-devel average: 0.0091820 MySQLdb 1.2.3c1 average: 0.0026477 oursql 0.9.1 average: 0.0007394 (MacOSX 10.6.2, MySQL 5.1.42, Python 2.6.1)
Well, it's a start. Now off for a walk!