Lambda the Ultimate

Has anyone seen an objective comparison, e.g., quantitative analysis, of whether unit testing will find the bugs an ML-ish type system will reveal? There are additional benefits to both approaches, and many claims and anecdotal experiences about these things, but I'm wondering if anyone has seen anything more objective?

For example, the DiamondbackRuby etc. guys might have something like a true positive and false positive rate on checking codes that were only unit tested.

Edit: Ask and ye shall receive (in a tiny part): a study of the bugs found by translating some unit tested Python apps into Haskell.