Good news, D..
Here's a plus for 32bit based on my experience last night. I forgot to lock the recorder (PCM-A10) and when I checked my levels via phone 10 mins into the first act, levels were maxed and I was hard brickwalling.
Had I been using a 32 bit recorder (like the Deity or F3), this would been recoverable. Luckily I only lost 10 minutes and none of the headliner.
True that if you had been using a 32 bit recorder, the input level would have been out of your control and low enough from the start (extremely low). However, you can run the A10 in essentially the same way by resisting the temptation to increase recording levels above whatever minimum setting is needed to avoid overload in the loudest recording situations you encounter.
Doing so will not be a problem as long as..
[snip..] the ambient noisefloor is significantly higher than the noisefloor of the MixPre at -18 [of your recording chain, in this case determined by microphone self-noise>recorder EIN] after I bring it up in post, so no cost to having the extra safety buffer [..snip]
SMsound describes classical/opera which has an ambient noise floor that is about as quiet as any taper-recorded live performance ever gets. The biggest problem for tapers is that not increasing recording level is a very difficult temptation to overcome! it's one that has become ingrained by habit as it used to be important but generally isn't any longer, 32bit or not.
The parallel switching ADC designs and very low EIN of 32bit float recorders help extend dynamic range sufficiently to fully accommodate very low ambient noisefloors that don't occur in concert taper situations. Most modern recorders that are not 32bit float can be used in essentially the same way as 32bit recorders for concert recording. I use a DR2d for classical recording, which when recording in 24bit probably has a real-world dynamic range of only around 18bit equivalent or so at best. I've never actually measured it. But as they should be, the noisefloor of the recordings I make with it are dominated by the ambient noisefloor of the hall. Even with an effective range of only about 18bit and the DR2d's input levels remaining the same all the time, I use only two different gain settings on the preamp upstream of the recorder- one for classical recording (determined by the need to keep the recording chain noisefloor lower than the ambient noisefloor) and one for very much louder PA amplified stuff (determined by the needed to keep the loudest SPL from clipping). Once those settings have been determined, there is no need to look at or worry about them again.
Using the Zoom F8 instead and recording in 24bit, I don't even need two gain settings.