Peopleware has some great data about how listening to music can lower distractions but also reduces problem solving ability. The upshot was that test subjects could plow through code while listening to music, but tended to miss optimizations and shortcuts.
I think Peopleware is a great book, but it makes bold claims with very little experimental data. I personally find this hurts the integrity of the book and would have preferred if the authors were a bit more humble.
The music vs silence point, for example, is based on a tiny experiment with a small sample size. The kind of conclusions that are drawn from it in the book are a stretch.
