Equal volume across a long track of backing music

[Trebor]

Ok. Given the main issue - that it is pretty even within any given section several seconds long, and the main problem is between differences many seconds apart - would this suggest normalization would still be a good first step?

Normalization, (RMS or Peak), applies a constant gain-change cross the entire track:
Normalization does not adapt during the track to the changing volume of the track.
Compression does adapt, as does AGC & Level speech.