DVDdoug:
A theoretically ideal Class D amplifier is 100% efficient and doesn't dissipate any heat. Page 15 of the datasheet shows that it's more than 80% efficient across most of the output range, so at 20W output, you'd be dissipating about 1W.
20% of 20W is 4W. Where are you getting 1W of power dissipation?
Also on page 8 of the datsheet, the second chart at the top shows that at 12V, with a 4 ohm load, the PD is 5W. I'm still unsure if that is for both channels at once or for each individual one.
Normal music & voice has a peak-to-average power ratio of around 10, so with 20W (undistorted) peak output, the average output would be about 2W.
Well, that's good to know, if true. That implies that at 16W output for the 12V into 4 ohm case, at 85% efficiency, you've got 1.6W of average power * 0.15 percent losses = 0.24W of power dissipation.
But what the other fellow said about the power losses being tied to the switching of the mosfet and nothing else seems to make sense, and even if the average power output were only 2W, the dissipated power would remain the same because all you're varying is the duty cycle to get that lower power output while the PWM frequency, and thus the number of times per second the mosfets switch and pass through that high resistance state, would remain the same.