if i have a wav and compress one mp3 at preset standard and another mp3 at 320kbps cbr stereo then recompress with preset standard would there be a quality difference?

Without a doubt. Recompressing any lossy format to a lossy format (regardless if it is mp3 to mp3) will always result in lower quality output than the input file.

can someone explain why?

in essence aren't all mp3's a subset of cbr 320kbps stereo? different codecs remove different information from the wav so it would obviously result in worse quality but i would think that 320kbps removes information that would be removed by preset standard anyway.

In a perfect information-theoretic view of a lossy codec that'd be correct, if you were using the same psychoacoustic model for all versions (the lower-bitrates would just remove additional information). However, in practice there's additional sources of error. One place is quantization error, which is greatly magnified when re-encoding. As a simple example, consider a 16-bit value that you're going to store with 12 bits. That might work okay; you just have a bit of rounding error. But now if you convert it back to 16 bits and re-quantize to 9 bits, that's a lot worse than a 16->9 quantization would've been.

An additional problem is that codecs cannot tell what's been introduced and what was original. For example, if an encoding produces artifacts (even really minor ones), the re-encoding will waste bits trying to reproduce those exactly, since it sees them as signal, which may end up magnifying them or making other artifacts worse. This is why you can re-encode a file lowpassed at 10 kHz in the first encoding pass and see the encoder trying to store signals over 10 kHz on the second pass.

There's numerous other problems, but these are the biggest sources of quality loss.

Ogg Vorbis is working on something more akin to the information-theoretic view of lossy compression with its bitrate peeling. The idea (very roughly) is that in the first (and only) encoding pass you identify information by various levels of importance, rather than just by "keep" and "throw away". Then you can "peel" the bitrate by progressively throwing out the least-important information. There's no re-quantization, and the encoding is all done on the original, so there's no problems with introduced noise as signal, so both the main problems I noted above are avoided. This is however very hard to do well in practice due to highly non-linear effects; it's very difficult to come up with clean subsets of data like that.

thanks. the peeling method being developed by vorbis seems promising especially for streaming.
would a stereo to joint stereo conversion without changing the bitrate be perfect?
what are the quantization levels for mp3? a right combination of levels would not give you quantization error. for example a 16-bit to 4-bit conversion is identical to a 16-bit to 8-bit to 4-bit conversion.

Your (initial) question (and similar ones) have been asked before. You might want to read some of the threads:

mp3 to wav to mp3 to wav same?

Re-Encoding quality loss

Probably you'll find more yourself using the search function. I haven't seen anything about this in the FAQ though. (Pio2001?)

Your question about stereo to joint stereo is actually interesting.
But I wonder, if there's an encoder/utility, that can analyze an MP3 stereo-file, and then do lossless coupling/joining of the two channels, and create a new MP3 joint stereo-file that sounds identical to the original MP3 stereo-file.

I don't know if this is possible though. However, it's not much of an issue, because:
1) Ripping in joint/coupled stereo is the best, so of you course, your files are already joint/coupled, when you ripped them originally - or maybe not
2) Even if such a utility could be made and create a valid MP3-file, I doubt a lossless stereo coupling of the original MP3-file, is going to create a significant size reduction.

yelloman,

I once had a similar belief, that lossy codecs might act like projection operators that just discard what we can't hear and retain what we can; members of the forum corrected me in this thread back in February. Some of their explanations were very good, and you might find them helpful too.