While I agree that the THD curves are weird, the curve is below 0.2% all the way until it gets at the max power. That is as good or better than many other amps. I highly doubt that any variations below 0.2% THD can be audible.
Vlad, you and I really seem to have two different perspectives when looking at the same info.
I will agree that 0.2% distortion is not audible as distortion, it will come through as less clarity to the listener. I would love to see a couple of examples where amp manufacturers have speced their amps at near 0.2%.
This has to be looked at in the context of how the specs on an ICE (or any class D amp) compared to more traditional linear designs. Many good linear designs are near flat on their THD curves vs freq. If you pick 1Khz as the test point, it is fairly representative of the whole frequency range. Now look at the 500ASP, it's speced at 0.006% THD at 1Khz at 1W. This spec neither reflects typical behavior across the frequency or power spectrum. Most linear designs will rate their amps at full power with 0.1% distortion but until reaching clipping, typical distortion levels are far below that number.
I am not saying that all class D amps are bad, but they should be compared to linear design in a fair and like manner. This is especially true if the manufacturer is taking a module of average performance is marketing it a a high end product.
As an example, look up the Hypex modules I mentioned earlier. Then look up Channel Islands Audio and Kharma. They both build monoblocks based on the same underlying technology. CIA sells its monoblocks for $2,300 per pair, Kharma for $6,800 per pair. Other than the very nice enclosure, what has Kharma done to deserve such a high price?