Dither *is* applied to the 24 bit signal, but I believe the only chance to hear it is after it has been applied and you are now listening to the 16 bit signal.
It is true, UV22 is akin to the bias signal of a tape recorder rather than the noise (shaped or not) called dither. But in direct comparisons I've tried, its effects are also audible.
"Noise shaping" does not push the noise into frequencies we can't hear. Take a look at the attached paper. You can see there is an increase in dither noise that is in the clearly audible region. This is true for POW-R2, L1-Ultramaximizer, L2. By comparison, POW-R3 is in a category of its own!
So, not only is noise shaped dither in the audible band, but the extra energy is some reasons exaggerates frequencies. As I said before, POW-R2 caused extreme sibilance in the recording I just completed.
After much study, I believe I am with Brad. I am arriving at the conclusion that plain old TPDF is best.
Trivia: I read up on dither. It comes from the British. Bombsights, being built during WWII, were found to be more accurate in the air than they were on the ground. The shaking of the bomber(dither)evened out innacuracies and made the sight better!