I really like seeing people's thought process as they decide to switch something major, but if I were doing this same task, I think I would have tried more on speeding up the sql solution before scrapping the db.

There are several ways you can calculate a rank, and some (e.g. self join) are inherently slow. Maybe the problem is as simple as not having the right clustered index? Or maybe sql dbs can do more optimization if you tell them to use their rank function instead of selecting out a rownum? Postgres 8.4+, sql server 2005+, and Oracle all have rank functions (one of the 'window functions' in the SQL:2003 standard http://en.wikipedia.org/wiki/Window_function_(SQL)#Window_fu... )