It somewhat depends on how much code targets DB specific features... if your data store is relatively "dumb", it's easy enough to abstract out ANSI queries with specific abstractions to keep the code very generic. I'd say that mysql is probably the oddest one out in a lot of cases.

The db/table creation is probably the one part that requires the most tweaking.

I have to admit, I always hated MS Enterprise Library, specifically the Data Application Blocks (I think that's what it was called). Mostly because it was a lot of abstraction for zero gain when you only targeted one database. The only time it was worthwhile was when I worked on an application targetting multiple backend databases. Even then it was a lot of work.