Certainly these workarounds aren't great and I hope you're able to figure out a way to normalize multiple clips to each other (emphasis on "to each other," just doing a batch normalize like echo and flanger operations will not produce the desired result. I presumed this was why the program doesn't do it.)
What I've decided to do in the meantime is quieten the outro music track to about the same level of the voice track, export, new sequence, import rendered file, normalize audio, then finally re-export losslessly, which goes quickly.
This introduces some fiddly problems and I really don't like the wait of importing twice, but I'll live.