Philip Maymin writes,

any set of regulations that attempts to assign risk to all securities in the same way for all banks would necessarily create some regulatorily-favored group of securities because the objective algorithm assigned too low an amount of risk capital to those securities purely as a result of statistical flukes and selection bias. Were each bank to do the analysis itself, their errors would be more r andom and less likely. Unfortunately, with a comprehensive system of regulations, identical errors pervade all banks.

Tyler Cowen points to this article today, although it was written several months ago. David Henderson mentioned it back then.