RMS is a measure of power, not of perception. You don't have to defeat RMS to defeat perceived loudness.

But, you're right that a playback device could take measures to defeat the perceived loudness baked in (though they'd get quite complex, as the tools for baking it in are extremely complex these days, and I don't actually understand half of them, despite having studied audio in college and worked in the industry). I don't know if it would actually improve listener experience to do so, though. Ending the "Oh shit that's loud!" and "Why is this song so quiet?" problem would be positive for listeners, of course. But, at what cost?

"Analog radio would still be mastered stupidly loud. I would assume final stage compression in radio broadcast is keyed on peak not RMS level, again please correct me if I'm wrong. But it's a dying medium anyway."

Analog radio (and digital broadcast radio, as they still compress, despite some of the technical reasons for doing so being gone, inertia is strong in broadcast; I have noticed that Pandora and Spotify do not do any sort of compression, however, which is nice...but also annoying when the playlist has new and old music as the difference can be massive and unnerving) was among the earliest adopters of various technology to make music perceptively louder. Aural Exciters (a very early salvo in the loudness war) were available in a broadcast targeted version from very early on, etc.

I haven't been in the broadcast industry in a long time, and I worked in television rather than radio (though the station I worked for shared a tower with several radio stations and another TV station), but I'm reasonably confident radio stations in major markets still tend to have the most modern "make it loud!" devices available. Loudness=listeners in radio. That's why the loudness wars are happening.