Jerry Goldman and Martin Katz published a rare empirical study of IRBs in the Journal of the American Medical Association in 1982. Here is a brief summary:
We submitted three flawed medical protocols with the expectations that the review boards would find the flaws evident in the proposals. But the boards, on the whole, did not find the flaws. Rather, they suggested changes that would, if approved, made bad research projects worse. This is the only major empirical study of IRBs and the findings have never been controverted with other evidence.