This isn't snake oil -- it's actually kind of a cool idea. The guy behind this is Heeb Thierry, who was also the mathematician behind the Anagram ASRC.
A better way for most people to understand it intuitively is that if you start with a 16 bit signal and have 24 bit output, if you limit your volume control to bit shifting, you don't actually need to add dither because you're not introducing additional quantization.
In the AES paper, they still do propose dithering at the last stage to preserve a constant noise floor, so any advantage to the system is a little more subtle than the intuition above (and requires more math), but you get the idea.
Where I think the company's marketing falls down is that it isn't true that most DAC ICs truncate after 24 bits these days, but if S/PDIF is involved somewhere in the chain, then that is true.