I uploaded the mp3 files from the source file again and this cured the problem - no more problems.
I also tested it by downloading the track from the empeg and reloading it to a new playlist - this also sounded ok.

So, I guess we put it down to the flawed/experimental normalisation routines in 2.0b3?
[I did play this track when I had 2.0b3 (std) installed so that may have screwed the normalisation possibly?].

Anyway, case closed for now.