I think it's more about the timeframe that the dynamics cover. Basically it's inter- vs intra-word dynamics. It is possible to get each word to about the same volume overal without completely destroying the dynamics within the word itself and loosing the transients of the consonants and whatever. It can be the same with other instruments also. Like take a drum track. Maybe one hit is really loud and the next is kinda quiet. So we can compress to bring the loud one down closer to the level of the quiet one, but we don't really want to mess up the internal dynamics of each hit. We want the attack to be as much louder than the sustain as it was, just the whole thing turned down.
This is very much why most folks prefer automation for that sort of leveling - basically because automation can go slower and be more respectful of those short term dynamics while still reducing the longer term dynamic range. There are a couple of ways to get something similar in a more automatic way, but it generally comes down to some form of compression with a relatively long integration time. Plugs like VocalRider and probably that Waves thing do this. I use ReaComp, set Attack and Release to 0, set the RMS time to like 500, and the pre-comp to 250 (actually, I set pre-comp as long as it goes and RMS time to double that, and end up with these numbers), ratio really doesn't need to be very high (often just like 1.1:1, though on vox you can sometimes go heavier), and the threshold and knee get kind if worked until the thing is always doing something when there's anything happening on the track. I usually switch on the auto-makeup just because it can be a lot to make up and the output slider doesn't really go that high. It works so well that it feels like cheating.
Course that's one step in the chain. Before that I usually have an EQ, often pulling out some bottom end, but almost always with and all-pass filter or two to "rotate" or "randomize" the phase of individual frequencies with respect to one another. This tends to make asymmetrical sounds (like many voices) quite a bit more symmetrical, which isn't quite the same thing as compression but can sometimes help make a compressor work more consistently, and just generally make the whole thing a little more full sounding.
Then after the long, slow, gentle leveling comp I'll often have a bit faster and more aggressive one to squish things down a bit more and start to control more if the intra-word dynamics like consonants that pop out or whatever. Sometimes there'll even be two getting progressively faster, with higher ratio but also higher threshold on each.
Which is an important consideration also. When we're talking about plugins - where we're not limited by the number of actually compressors we have in our rack - we can use as many instances of any plugin as necessary. Eventually have to worry about CPU, but... Slap as many damn compressors on there as you need! Most of the time several compressors each doing a little can get better results than one trying to handle it all at once.
Don't know if any of that helps...