Singers whose dynamics vary with pitch like that are good candidates for compression - that's when it should be used. If there's any accompaniment going on, the voice is unlikely to sit sensibly in the mix without compression. When mixing, you apply the compression just to the vocal track, not the whole mix. That is a far cry from extreme limiting across the mix of already compressed material so it hits with -0.1dB of full scale every second of the music.

This has been suggested before (and slightly more on topic) : music should be delivered in an uncompressed form and playback equipment (especially in cars) should have a compressor built in to keep the music at a constant level above road noise. It's already quite common for car radio/CD players to have a volume level tracking the speed of the car; compression during playback is further refinement of the same idea. Then you wouldn't need two copies of everything.