Ok, well it really looked as if you were talking about mastering…
But again you have to be clear what you’re talking about. If you’re literally talking about headroom during mixing then the channels have a tremendous amount of headroom, far higher than we need, because of floating point processing. If you’re talking about individual plugins distorting because they’re programmed to then that doesn’t ‘follow’ 0dBFS, it ‘follows’ whatever reference nominal level the programmer has decided on. That’s most likely going to be an average level like -18dBFS, not peak level.
And if you’re talking about the final output on the master then we’re back to talking about what to deliver to the mastering engineer and that’s different from the mix. The standard advice to not clip the output always applies.
Using True Peak metering you can see they all go somewhat above 0dBFS, but sample peak is limited to exactly 0dBFS and can never be higher. That means that in order to get the signal louder you have to do something other than reaching 0dBFS on individual samples. If a super-loud metal master has samples that reach 0dBFS and a master of Adele reaches that same 0dBFS then what creates the difference in loudness isn’t those peaks, it’s “the rest”.
As a matter of fact I’m willing to bet that if you take for example a moderately loud master from say the year 2005, maybe something like a Paul Simon album, and put that up against a super loud metal master, then you can probably lower the metal master quite a bit and it’ll still feel (and be) louder than Paul Simon’s record, even though the metal album’s peaks are now solidly below 0dBFS and Paul Simon’s are still peaking higher (sample peaks).
That’s because actual loudness isn’t determined by peaks but by average.
(PS: Again; lowering all of the audio in a file obviously reduces the loudness of the file, but the loudness of “the mastered music” is in a sense the same. There’s a slight distinction.)