This thread wasn't actually about RMS compression, but rather the exact opposite: peak limiting. But you know what's super funny? The best, most transparent peak limiters do exactly the same thing that I was talking about, only with much shorter times and much heavier ratios.
If you go through sample by sample, compare each to the threshold and if it's over turn it down some amount based on the ratio, you get distortion. This is what every saturator, clipper, overdrive pedal, and guitar amp does to get that sound. In a compressor or limiter when this happens, we call it "ripple distortion" because the signal in the detector path is not smooth, but "rippled" by the audio signal. We cure that by slowing down the detector. Most of the time, that's done with the attack and release controls. These are usually very much just low-pass filters on detector signal that take out those fast wiggles so that the gain adjustment happens more smoothly. RMS window does the same thing also. But it's a tough balancing act, especially at limiter-type ratios. You can only get so fast before it sounds like distortion, but if it's not fast enough you miss the quick transient peaks that you're trying to take.
Worse yet, you're still adjusting gain based on the average of the past however many ms. You're still turning down today's volume based on what it was yesterday. You're always going to miss the leading edge of a quick sudden spike. You're always trying to play catch up. This can actually make things much worse, as if the limiter kicks in hard in the middle of a quick spike, it basically makes that spike even quicker!
So a lot of the really aggressive brick wall limiters actually have a short RMS window and a lookahead halfway through that. They see the spike coming and start turning down a little early so that when it hits, it's already down as far as it needs to be. ReaComp can do this too.
Unfortunately, that style if limiting isn't super reliable. For a number of kind of weird mathematical reasons, it's difficult to predict exactly how much attenuation any given input will really get, so that even with infinite ratio, some peaks can still go over the threshold, so your ceiling isn't really a ceiling.
That's when you just stick a sample-for sample saturator/clipper whatever so that you know for absolutely sure where your loudest peak will be.
...until you turn on over sampling...
...or it hits the reconstruction filter on the way out the DAC...