I was going to comment last week on the "molecular evolution doesn't reverse itself" paper -- it's a very clever use of informatics and the comparative method, and I saw a talk on it last year. Carl Zimmer describes the story:
Based on these results, Dr. Thornton and his colleagues concluded that the evolution of the receptor unfolded in two chapters. In the first, the receptor acquired the seven key mutations that made it sensitive to cortisol and not to other hormones. In the second, it acquired the five extra mutations, which Dr. Thornton called “restrictive” mutations.
These restrictive mutations may have fine-tuned how the receptor grabbed cortisol. Or they may have had no effect at all. In either case, these five mutations added twists and tails to the receptor. When Dr. Thornton tried to return the receptor to its original form, these twists and tails got in the way.
Still, it's not a demonstration that molecular evolution couldn't reverse itself -- it's just that the inverse series of molecular changes is not adaptive upon reversing the selection pressure (in this case, for cortisol specificity). That's a restrictive meaning of "reverse". The receptor might (or might not) respond to a reversal of selection pressure with a different series of substitutions. Or it might follow precisely the same sequence in the absence of any selection -- although like any precise sequence of random changes, it would be infinitesimally likely. Zimmer's story gets this nuance out of the authors, making it a better description than any others I've seen:
If this molecular Dollo’s law holds up, Dr. Thornton believes it says something important about the course of evolutionary history. Natural selection can achieve many things, but it is hemmed in. Even harmless, random mutations can block its path.
“The biology we ended up with was not inevitable,” he said. “It was just one roll of the evolutionary dice.”