CD Ripping Uncertainty Principle?

I was recently playing with a Mac OS X port of cdparanoia, and in the
course of my testing, I discovered some interesting things about
cdparanoia; things I wasn't very pleased to discover.  I decided to rip
a CD on both OS X and Linux and compare them.  I used cdparanoia with a
SCSI Plextor 12/10/32 for both trials.  The rips went smoothly, and I
got no errors or warnings of any kind.

Well, out of 9 tracks, 8 md5summed the same.  The 9th didn't.  WTF? The
WAV headers were the same, the data alignment was the same, the file
sizes were the same.  So I whipped up a little program to do some binary
diffs... turned out there were a couple of dozen samples near the
6-minute mark of an 8-minute track that were just DIFFERENT.  Not by
much, never by more than maybe 1% of the total dynamic range of the data
type.  Probably not even audible.

So I decided to do ANOTHER rip on Linux, same setup again, and added it
into the mix.  SAME BEHAVIOR! A bunch more samples in the same area of
the disc got added to my collection.  Sometimes the results of the the
third run agreed with the first, sometimes the second, but always with
one or the other. Strange.  So I did another run, this time still with
cdparanoia but now on a laptop running OS X.

More of the same! There are just a large number of samples where any
given rip will result in getting one or the other of two possible
values, and there doesn't seem to be any rhyme or reason as to which one
you get!  The only difference with the 4th rip, done with an ATAPI
drive, was that I actually got some feedback from cdparanoia... I got
the ":P" face and a "+" in the progress meter at the trouble spot.
Still, "unreported loss of streaming" doesn't sound as though it should
result in quantum effects when reading a CD!

I did two more runs, including one with iTunes, and got more of the
same.  The only difference between the files was that tracks ripped on
different platform setups gave me differently-offset files, which I had
to hand-align in order to analyze. At any rate, I listened carefully to
one of the rips and I couldn't hear any problems at all.  <shrug>

At this point I have no idea how common this sort of thing is.  I've
only tested one CD (a commercial one) and it only has this one problem
area maybe a second long or so.  Maybe lots of CDs do this, maybe it's
quite rare. I have no idea.

Does anybody know why this might be happening?  I've been using
cdparanoia for about five years now, but this is the first time I've
ever attempted a serious analysis of its output...

Thanks,
Alec