Didn't like his comment near the beginning about the digital format supposedly not being able to reproduce the continuous analog waveform; doing that is the whole idea behind digital sampling.
I think what he was trying to say is that an increase in sampling rate may help a CD. Building the disc out of esoteric matterials: not so much.
But yeah, a well produced CD can sound better than a hot, over-compressed DVD-A (Flaming Lips, At War with the Mystics; I'm looking at you).
More samples per second can help, not just extend the upper frequency range beyond what dogs can hear, but allow a more accurate representation of upper few octaves. Imagine if you will, a 48 dB, 11 kHz sine wave. That's a full octave below what Nyquist says can be reproduced by a 44.1 kHz sampling rate. But if the samples are not taken right at the peak, trough, and zero crossings, you'll not be representing the full energy of the original analog waveform (too, @11 kHz that sine becomes a triangle wave even if the samples hit right on mark).
In real life this doesn't happen. No one records an exactly 11.025 kHz sine wave, and even if they tried there'd be clock drift between the source oscillator and the ADC. Also all mastering is done these days at least at 48 kHz, and many times 192 or higher. This allows the mastering software to get a bigger overview of what the original waveform looked like, and antialias the samples so sine waves don't lose their energy if they don't hit right on the mark. But another word that pretty well describes "antialias" is "smear"; sure the energy is now preserved, but that comes from averaging it into neighboring samples.
While most adults can't hear up to 20 kHz, they can hear 10 kHz, and the 441000 sampling rate can still have serious aliasing artifacts down as low as 5 kHz. Antialiasing in the CD mastering step can help, but does lose detail; there's no way around it. Preserving more of the original samples of a 384 kHz master recording, onto a 96 kHz distribution media for the can help a lot.
ADDITIONAL: I'm big on double-bind tests to prove to myself that I'm not just imaging things. I've actually conducted an experiment with my project studio gear. I have a multi-channel sound card with 24-bit 192 kHz ADCs and DACs. To the ADC I connected a mic and pre-amp, and the DAC outs were connected to an analog mixing board where I was listening through it's head phone amp. The mic was just picking up environmental noise from the living room: people moving about, talking, cars through an open window, etc. I was in the bedroom, headphones on, eyes closed, while my brother changed the sampling rate of the sound card (both the input and output clocks were changed together). When stepping up and down I could tell him, when the change was made and what direction (higher or lower rate) he went with almost perfect accuracy up to the 48 to/from 96 kHz change. Sometimes I'd detect the change from 96 to or from 192, but most of the time I'd not know he'd even touched it. What I could not do is identify one sampling rate from another at random. Only the improvement of going higher vs. the loss of going lower.