I think that <0.05 THD at 100kHz is useless voodoo, since any of us are lucky to hear anything at 20kHz let alone 5 octaves higher than that, and your CD player isn't producing anything at that level. That doesn't mean the tweeter isn't great, just that that statistic is useless.
Two octaves. But, in any case, nothing in life is
ever that simple.
It is true that a person with normal hearing cannot hear anything if you play him a sine wave tone with a frequency much above 20kHz. However, it does
not follow from this that the ultrasonic components of the signal are useless. It has been fairly well established now that the ear is actually sensitive to "transients" which include ultrasonic components.
What is a transient? An example would be someone hitting a snare drum. If you model the sound pressure level following a snare drum hit, the signal you get is very close to vertical: it goes from nothing to about 130dB in almost no time at all. If you filter out all components above 20kHz then you cannot represent a rising edge this steep - it can't rise any faster than a 20kHz sine wave does. And the ear
is actually sensitive to the difference between a 20kHz sinusoid rising edge, and the much steeper rising edge that you get with (say) a snare-drum hit. If you extract the component above 20kHz and
repeat that pattern,
then you don't hear anything - but the initial attack is audibly different without the ultrasonic portion of the signal.
Now, it has to be said, if you're listening to a music CD, it won't contain anything above 20kHz anyway, so you'd have to be listening to something like DVD-Audio or SACD for steep rising edges like this to actually be preserved in the recording.
The other point, of course, is that a tweeter capable of very low distortion at 100kHz will be producing even less in the 5-20kHz range, and even very small levels of distortion in that range are very audible indeed. One of the comments I've heard from people who have head the diamond tweeters in action is that they're one of the few speakers to really reproduce human speech perfectly - especially consonant sounds and particularly sibilants ("s" sounds). A spoken "s" has a lot of frequency components up in the high-tweeter range that a lower-quality tweeter messes up, and makes the speaker sound as if he or she has a very slight lisp.
