Yeah, Tom Lund is a smart guy, background in broadcast audio.
I think first you have to (mix and) master properly. I do not compress much on strips, unless the musician in question (me) can't keep a consistent level. With samples, that is never a problem. If you get the mix right then you can master and keep the level even more consistent, and apply techniques to get a 'louder' sound, if you can tolerate the tradeoffs.
No buss compression, either. I use about 3 db of compression at about 2:1 or 2.5:1 in the Multipressor, maybe a little on certain individual instruments (maybe too much parallel comp on snares). Then, the Adaptive Limiter can get you about 6 dB of loudness before it strangles your dynamic range to death. I do none of that to chase 'loud', I do it to chase 'glue', because that makes the sound better.
But even if you are (quite stupidly, and I know you are not) chasing 'loud', I have found that if I want my tunes to match the level of iTunes, which I use as a reference, I need to normalize in the final bounce.
Normalization has a bad rep, but that's for other reasons. If you are putting out a final copy, you don't need headroom any more, so normalize and eat it up. It does not change the sound at all, not one 'bit' (pun intended), all it does is find the loudest sample, raise that to value 111111111111111111111111 (or 1111111111111111, if going to 16 bit) which is dBFS 0, and then raises the value of every other bit correspondingly. It's a simple arithmetic function. So all it does is make it louder (including the noise floor) and it is completely transparent, so not problematic in any way. Just don't normalize before doing it here.
The only secret here is that all of those mastering concepts only work properly if the song is mixed with proper levels in the first place. And if you can get the levels right in tracking, you don't need much compression there (unless the goal is not consistency but altering the sound character).
And if you can get the mix right, you really don't even need that much compression or limiting there, or in the mastering. But if you do that right and the levels are consistent in the mix (no notes sticking up) that means normalization as a final step (before dither) can bring the entire final level up more than it could if the mix levels were inconsistent. The highest peak is the roadblock that stops normalization from going higher, so if the peaks are consistent, the end product will be louder.
Look at the waveform directly on the mix. If there is an occasional peak that sticks out above the rest by 3-4 dB every few measures or so, those peaks will keep the normalization from raising the level where you want it, by that amount. Once that waveform looks (and sounds) consistent, with no outliers sticking up, normalization works wonders.
(of course I am speaking about 'loudness' in the sense of comparative 'volume', and not in the sense of 'perceived' loudness)
I find that a lot of my songs have different levels coming out of the master bus (which sort of alarms me since I gainstage religiously), but if I normalize on the final bounce, everything sounds the same as stuff in iTunes (just a lot better. Ahem!)
