Also ... just to nit-pick, it is not the sampling rate that we are
discussing here. That is still going to be 44kHz (I think that's the
norm). It's the compression that is the issue (as MZ was mentioning in his
reply) 128 is a much HIGHER compression then 320 which is why the files
are smaller. Someone along the way told me that at high compression the
range of frequencies is cut off (something like 30Hz - 18,000 Hz) and
similar frequency patterns are simplified to conserve space. I am not an
expert on the encoding of MP3, so all I can do is throw this out for the
wolves to chew on. With lower compression (192, 256, 320) not as many
corners need to be cut ... hence the better replica.

High frequencies take more "bandwidth" (not really the appropriate term
perhaps) to encode than lower frequencies. That's why it's typical to see
the signal low-pass filtered rather than high-pass filtered. Some encoders,
like LAME, allow you to adjust the crossover point. The "alt-preset
extreme" option sets it somewhere in the 19kHz range, which is more than
high enough. For what it's worth, many folks can't hear higher than about
17kHz anyway, and even if you've got superb hearing you'll have trouble
getting that in the car. But, as you said, hard drive space is so cheap now
that there's no sense in cutting corners.

In fact, some folks are simply transferring their entire collection to wav
thereby leaving it completely uncompressed - screw mp3! Mp3 was useful when
hard drive space was at a premium. And, for the small portable players,
which still tend to have limitations on space, you could always "encode on
the go" - that is, you simply encode it from wav to mp3 when you transfer it
onto the portable device. With CPUs these days regularly over 2 or 3GHz, it
takes no time at all to batch encode.