> How long does a song have to be before the accumulated errors add up to that many microseconds?
  Part of stereo sound perception is from the delay between the time a
sound reaches your two ears.  At 1000 ft/sec, that's about one
millisecond.  Suppose two waveforms are controlled by two separate
computers, whose clocks drift apart by 1 minute/day (I've had worse).
They will differ by one millisecond after 1.44 seconds of elapsed time,
so one could imagine the sound of two instruments with one of them
appearing to be running from stage left to right and back again every 3
seconds.  Noticeable? ;)