With the growth of the PLoS journals, there's recently been a rise in discussion of alternative forms & variants of scientific peer review. In this context, we'd like to raise the big question: What good is peer review, anyway? To do so, we'd like to use as an example, not a journal article, but rather a book.
In 2002, Stephen Wolfram, founder of Mathematica (official corporate motto: "No, we're not MatLab."), self-published a big book, modestly titled "A New Kind of Science". The following appears on page 443 of that book:
"… [a] systematic decrease in randomness … would certainly in principle be possible … But somehow it seems that [suitable] initial conditions … never actually occur in practice."
"Never"? In fact, the appearance of order out of apparent randomness, due to carefully crafted initial conditions, occurs (conservatively) billions of times every day, in the world's MRI scanners, as NMR spin echo formation is essential for much of MRI.
What else in the book would have surprised (and been corrected by) peer reviewers, we cannot say.