Oh yea, I forgot to come to the defense of my buddy Doug.
Someone wrote, as a challenge to Doug:
Nyquist theorm is only to detect a signal and not to reproduce.
That is false. The Nyquist theorem is about exact recovery of a continuous signal from discrete samples (in particular, from fast enough samples of a bandwidth-limited signal, the sampling going on forever just as the signal does).
The recovery is exact mathematically. This is a mathematical result, and it tells us something interesting, practical, and approximate-but-very-close about the real world too. Very nice. Very important.
Someone else pointed out, correctly, that there is a problem at the EXACT folding frequency B, and that is correct too. But this is a well-known little side point, nothing more.
Back to the main point. In the real world we don't have the exact values at the sample point, we have to quantize them (into 16 bit or 24 bit integers). That means we have lost exactitude. Also, computing circuits cannot exactly compute the mathematical (sine) functions needed for the exact interpolation, can't do division exactly when irrational numbers are involved, etc. ... so more error.
And the model is wrong, since audio signals cannot really be bandwidth limited in the mathematical sense since they are time-extant limited, as I described above.
None of this matters. What matters is how remarkably close to the real world the theoretical Nyquist result is (and it's not really Nyquist's, but you can check Wikipedia for all that). DACs are great, sampling works. In his paper Lavry says maybe 44.1 is a little too low, but 96 borders on overkill. That seems right to me. This does not mean that upsampling is useless -- just the opposite, it is a neat first step in some interpolaton algorithms. I find some DACs sound better if I upsample externally to 96 first, even if the DAC upsamples on its own -- it all depends on the algorithm used.
An analog reproduction system makes all kinds of errors too -- tracking, etc. Nonetheless, the best analog systems sound great. So do the best digital systems. Poor digital systems seem to introduce errors that annoy people a great deal, and that's what gave CD's their initial bad rap among audiophiles.
Maintaining a first-class analog system takes a lot of manual work. Especially when using a computer audio system, maintaining a first-class digital system is almost no work. Both can sound fantastic. Stick-shift vs automatic transmission ... is that a good analogy? I never had a good, racing-quality automatic and thought a stick-shift in any high-performance car was necessary, but then in the early '90s I got to drive some automatic Fords with real performance boxes and I realized I was wrong.
I'm getting a high-end turntable so I don't have to re-buy all my LPs, but everything I buy now is digital, even if vinyl is available.
Listen, and do what seems right to you.