Although the tracks have been normalised to have the same average loudness, the more aggressively compressed tracks that have less dynamic range will still sound louder at this lower level. I don't see how what YouTube is doing is going to help.
If the lower dynamic range tracks sound louder even after normalization then the normalization algorithm is flawed. ReplayGain weights the energy by frequency to better match perceived loudness, and it does a reasonably good job. Other algorithms might do an even better job at matching human perception.
Those two things have a pretty similar loudness(A-weighted RMS) and yet the compressed one sounds louder than the measured difference - ~3dB A-weighted which is not obvious to perceive untrained.
As a test - download them both, amplify the Uncompressed one at -2.7dB and listen to them again. They have the same A-weighted RMS yet the Uncompressed one sounds louder.
As I understand it, tracks with more dynamic range will have louder loud's and quieter quiet's, while those with low dynamic range will have less of either.
That's true but most people are used to a smaller DR(up to a point of course but that point is way out) and still prefer the lower DR even after normalization.
