it's gain structure.
there are many different situations where you need to boost / cut
and they each have a set of tools and logic behind the choice.
digital is different than analog in this case because the "noise floor", apart from actual noise, also suffers from aliasing (phantom frequencies) when it's boosted.
once you loose the information on the bottom bits it's gone in the chain and further boosting can cause aliasing. thin sound with unpredictable harmonics to my ears.
try attenuating , recording , and playing back while boosting a long fading reverb tail to hear this clearly.
also, you still need headroom on the faders and I wouldn't recommend having many of them above the unity line regularly.
if you set the gain structure correctly, you should arrive at a point where the fader positions represent the balance.
it's one way and there are others, but it makes good sense if you think about it.

for example from analog, if the tape return signal of a track is too weak, you adjust it with the line trim (equivilant to track trim or normalize) so you have more freedom and fader resolution.
another example, if the eq attenuates your signal too much, you adjust it with the eq output.
visa vera, if it boosts too much, cut it at the input to avoid cliping inside the plugin. (if you clip the signal before the output and then lower it there, you get a clipped lowered signal).
compressor "make up gain" is self explanetory.
another point is not to do 2 inverse calculations,
such as cutting the signal very low at one point
and then boosting it way up right before the summing bus.

summing it, the 2 main pifalls on either end are aliasing due to quantization errors (soft end, emphasized by boosts of low level signals)
and of course digital distortion caused by cliping anywhere in the chain.

that's maybe enough for examples because there are almost endless situations.

in the end most important is to listen. not all algorythms are the same, so the one that sounds best to you is the right one to use.

I'll try in different words.

in a nutshell, quantization errors are more of an issue with softer levels
because they actually "produce" content that wasn't there to begin with,
since the resolution in the bottom end is quantized to fewer bits,
leading to low level, random (erronuous), points above and below the zero line, which translate to random phantom frequencies.
if you fall down low enough in level along the processing chain and boost back later
this is clearly audible, like the experiment I described.
bounce a reverb tail on -60 and boost it back up, nothing to do with noise floors and hiss, a purely digital phenomenon.
that's why I recommended maintaining a healthy gain structure.
of course if you have 48 bits at your disposal internally you can go lower than you would with 16 or 24 before this effect kicks in, for the reasons I mentioned above.

this particular side effect of quantization errors doesn't happen with high level signals because they're far enough from the "zero" line,
and what happens there is a different issue than what I'm talking about.
avoiding unnecesary proccesing is always true, and since were talking about gain structure in general,
I think understanding what happens at the extreme ends of it is important to the discussion.

of course instead of writing all this I could say "don't go too soft and boost back up cause it'll sound like tin foil".