My comment to another comment following Mark Waldrep's interview of John Siau:
The 6dB SNR would apply if there were no filtering, it is the wideband delta sigma noise up to the nyquist frequency (1.4mHz for DSD x64). While it might seem at first glance that DSD has wider bandwidth than any PCM, because of the astronomical sampling rate, that turns out not to be true, because of the huge need for filtering the high frequency noise, which doubles in each higher octave for delta sigma modulation. I applaud you for doing critical blind A/B testing and honestly reporting what you have not heard. More people should do that before claiming huge benefits to technologies like DSD, and therefore saddling us with cumbersome and restrictive formats that do not allow user digital signal processing, something I do in all my systems, and the digital signal processing has immediately obvious audible benefits (I do dsp in 24/96), unlike these differences between digital formats, which I feel are overhyped.
As for myself, I have not done so much blind testing, but it is my belief, from JAES published research, and casual listening, that supersonic frequencies ARE at least somewhat important. My best systems use super tweeters with response to 40kHz. So I do care about extended frequency response and low noise to at least that point. Furthermore high frequency loss and noise in an audio system is cumulative, so it has been reasonably argued that any given component should have response well above 20kHz, and low noise out there as well. I think it is wonderful to have audio files in 24/96 and 24/192 for this reason, among others.