It would be interesting to see this work expanded to include some of those factors; Namely indexing and archiving. Bad data models seem like a whole other beast, but indexing and archiving should be possible to generate good configuration suggestions for if you had a full history of DB usage trends along with what you were hoping to optimize for.