Once upon a time, as an intern, I had the 'fun' task of optimizing the performance of a stored procedure written by someone with a math phd. It was more than 6 pages when printed, and took more than 30 minutes to run (it was used in billing), and had no tests.

Ended up rewriting it in native code, and it run in less than a second. Most of the work was proving it produced same results... and writing and documenting test cases, so next person wouldn't have to go through that.

After that experience, I have generally avoided putting a lot of business logic in SQL...