For generating the files, maybe it's a better idea using triangular dither instead of strong ATH dither, because the later can work worse if further processing (such as digital attenuation or eq.) is performed when listening to the files. Also, there's no need to use dither when converting from 16 to 24 bit.
About 16->24 conversion: Dither or not should be equal, so no need to change this - or...?
So ... In your statement there's a "maybe" and a "can". This means to me either you're a polite person or it's hard to tell if noise shaping should be used or not.
Given someone who uses 24/96 audio for storage/playback probably uses high quality equipment - will this decrease the danger of noise shaped dither giving worse results when processing is applied?
AFAIK it has been tested that hf inaudible tones don't intermodulate inside the ear and cause audible tones, but I could be wrong. So... I don't know that much.
It was very short, I just tried 6 kHz and 18 kHz and heard nothing else than 6 kHz. That's all.
Max Sample Value: 27755.65Peak Amplitude: -1.44 dBMinimum RMS Power: -70.73 dBMaximum RMS Power: -8.69 dBAverage RMS Power: -22.1 dB
This disc should be of great help: http://www.chesky.com/catalog/body_catalog...0010&CATEGORY=2 - unfortunately I can't find it at the library. Heck, I'd buy it if I could find it at all in Holland...