Here, let me put this out there again:
I'm going to start up matlab. I'm going to generate a 20khz signal sampled at 44.1 kHz. I'm not going to run it through the anti-imaging filter, yadda yadda, just plot the 20kHz signal with two different phases, one 1/128th of a cycle (well under one sample interval to say the least) phase shifted.
See if you can find the difference in the two (different colored) plots, m'kay? Every spot you see blue is the difference. I think it's rather visible.
The code that generated this is:
w=2*pi*20000/44100;
>>
>> x=sin(w*(1:64));
>> plot(x)
>> p=pi/64; %that's 1/64 of half a cycle
>> y=sin(w*(1:64) + p);
>> hold
Current plot held
>> plot(y)
Now, yes, it has "beats". Those "beats" are with the image of the 20,000 Hz signal at 44100-20000 Hz and higher images. They go away with the anti-imaging filter, too. I have a nice graph out there somewhere showing this.
The period of a 20kHz sine wave is 1/20000. The time shift shown in the graph is 1/(20000 * 128) = .39 microseconds, rather smaller than the 22 microsecond sampling interval.
Furthermore, you can MEASURE a difference down to thereabouts of 1/(2*pi*20000*65536) seconds, give or take a few nits, and below that with additional averaging.
As they say, Q-fscking-ED.