A couple of points to remember: RMS is a way to average a sine wave. The RMS voltage coming out of the wall in most U.S. homes is 120 VAC, but the peak voltage is about 170V. That can be measured across a single wavelength.
Peak versus average power is different--it's how the amplitude (really: area) of the loudest wave (usually a very short period of time including however many waves it includes at the frequencies involved) compares to the average amplitude of all the waveforms in the music.
So, the loudest signals in the music can be measured and reported as peak or RMS values, and the average signals in the music can be measured and reported as peak or RMS values. It seems to me run into trouble when conflating "peak" with "loudest".
When evaluating continuous sine-wave test signals, those values merge.
Heatsinks and continuous ratings and so forth are rated for the average signals continuously and for the loudest signals over some shorter period of time. That's different from the amp's ability to fill out the full waveform of the loudest part of the loudest signal in the short term.
In the amateur radio world, transmitter amplifier power is rated as "peak envelope power", which is defined by the area of the waveform being produced. But they are also rated at that peak power for "intermittent amateur service", meaning that they assume peak power won't be demanded more than some percentage of the time. Amps for morse code running at full power are more demanding than amps for the voice modes--a morse key down is like those test tones, but the morse key is only down about 25-35% of the time in an actual CW conversation. The digital modes can be worse. Voice modes only demand peak output once in a while while the operator is speaking, even if the operator is using limiters to increase average loudness (amateur radio has its loudness wars, too), but presumably the operator is only speaking some fraction of the time, notwithstanding the apparent unwillingness of some operators to listen occasionally. Transmitters for continuous service, such as broadcast stations and (even worse) continuous digital transmission) are built quite differently even at the same rated power. Duty cycle is important. It seems to me that many Class D amps are built for a reasonable duty cycle for actual music, not for continuous full-power output.
I like seeing power plotted against distortion and noise at an important selection of of frequencies--I think that tells me how an amp will perform in the real world.
For me, "harsh" amps are those that distort on the onset of sounds in the treble region, requiring a finite period to stabilize on the correct wave shape. This seems to me what undermines a clean "sssss" attack to avoid a "sh" distortion. But that is arm-waving. I have amps that do that, when driven to their limits (or when malfunctioning, or when playing LP records that have been overplayed or ruined by a past crappy record player). I also have amps that compress output when it approaches the amp's power capability, which dulls the loudest sounds. I like having an abundance of power because it means I never have to worry about some transient peak finding the peak power envelope and going non-linear.
Now, to the topic: My current amps are Class AB amps of good design, reputation, and measurement (by the standards of a decade or two ago). They are rated to produce less than 0.09% THD at 125 watts into 8 ohms from 20-20K, with a dynamic headroom of 1.2 dB. The dynamic headroom describes how much louder the loudest signals can be without running out or the ability to fill out the waveform. An amp of noticeably greater power would need to be in the 250-300-watt range (into 8 ohms) RMS continuous, it seems to me.
But my old B&K Class AB amps have a signal/noise ratio of well over a hundred dB, which means that they are effectively transparent for me. But I do notice a very low hiss from them when they are on but no music is being played--I have to be within a foot or two of the tweeters to hear it. A Hypex NC502 amp would certainly be cleaner in terms of measured distortion, but the only noticeable improvements from my perspective, and in my ears, would be 1.) the greater power (maybe), and 2.) the absence of that quiescent hiss. I would do it for fun, and for efficiency and space, but it will not otherwise change my life. That's why I haven't done it yet.
Rick "finding it ironic that a few people who praise 20-watt SET amps complain about the way Class D amps are rated" Denney