This article shows the measured crest factor of some representative pieces of music. The conclusion is that, for a “typical” home audio setup, peak power in excess of 500 Watts may be required for an average power level of only a few Watts, if the goal is clean (i.e. unclipped) audio.
Note that heavily compressed audio has much lower crest factor (i.e. headroom) requirements. However, the average listening level is typically much higher, so the peak power requirements are still comparable.