I'm absolutely sure that I said something that can be boiled down to "In fixed point it doesn't matter much as long as you're not clipping, and in floating point it doesn't matter at all." In fact, the whole point I've been trying to make is that as long as there's not too much noise and there's not any unintentional distortion, it ultimately doesn't matter at all. If your file meets those criteria, and your Mastering Engineer says any god damn thing about your absolute levels, they're a dick and you should move on.
I'm not actually saying you should push your peaks all the way to 0. I'm saying you shouldn't NOT do so just for the sake of the next step in the process. Floating point files allows us to not have to worry about it at all.
Now we're going down the rabbit hole.
I've got my monitors calibrated such that when the knob is at 12 oclock, nothing in the analog chain is even close to distorting and an RMS level of around -15dbFS creates about 85dbSPL at my ears. It's not as loud as I want it to be, but it's actually just about right for a standard level. I turn it down sometimes. I crank it up sometimes. But I know where I'm at and it's about right when I turn the knob "home".
If I have a mix with a 15db crest factor, and the peaks are hitting -10dbFS, it's going to be pretty damn quiet at my listening position. So, like, am I supposed to turn up my speakers 10db just so that my ME can have some room to work? That's absurd. What happens when something actually does hit 0dbFS?!? These things happens sometimes. You hit the wrong button or unmute something unheard, or turn the wrong knob on some plugin. BANG!!!!!
Then remember that in my case (this thread is about "Mastering Software") "Mastering Engineer" is literally spelled "ME" and then what? I recalibrate my monitors when I move to the mastering stage? Nope.
So then I'm mixing, and at certain point in the process I drop some master bus processing on it. I know I'm going to add some glue and some squish and some saturation at the master stage, and I want to get some idea of how that's going to work, and then adjust this mix to best "survive" that treatment. But what actually happens is going to depend on the rest of the material that's going onto the album, so when I go to render, I'll bypass that whole chain. And since I've already got it up close to the actual honest to goodness FS levels I'm shooting for at the end, well some of the peaks are probably going to go over 0. So I render to 32 bit floating point. Then the adjustments at mastering are dbs rather than 10s of dbs.