going by a graph of sample volume also shows a single dimension of a multidimensional object – akin to judging, say, a photo of a 3-dimensional sculpture. looking at sample volume does not account for how we hear volume, which is frequency-dependent; limiters, whether analog or digital, apply a lot more than just volume compression and makeup gain. each has a sonic character in the frequency domain as well.
that's why you get comments about better separation and clarity – if volume and compression were the issues alone, you wouldn't hear differences that are clearly frequency dependent (cymbal definition through the wall of guitar). it's also why it's woefully inadequate to compare it to keeping the clipped master and simply backing up from the speakers: this can't achieve the same result because the directionality of the speakers and the amplification of the frequency spectrum changes with distance, and even if you did account for this somehow, the source is fundamentally different because the speaker analogy only changes transients because of where you hear them from (and how the DAC interprets the samples that have been clipped off), rather than the HDTracks master having the unclipped transient intact.
every song definitely had a compressor on the master bus, which is why they still sound like a glued wall. those would be intact on the final mixes, so if the HDTracks source is actually pre-mastering altogether, it wouldn't avoid that processing.
also, the standard is 24bit for recording and mixing. some places may have moved to 32bit floating point but 24 more than accounts for quantization error, so using a multiplier is generally adding unnecessary file size without particularly needed benefit.