Hi. I just noticed the technical nature of this thread, since I don't follow Zoom products in general. A few comments if I may.
First, neither floating point data storage nor gain-ranging by combining the output of two converters per channel is new in digital audio by any means. The combination may well be new in a consumer recorder, I dunno. But these techniques haven't been all that widely used in professional digital audio up to now, not because they're "so advanced," but simply because neither has offered sufficient advantage so far.
Neumann's digital microphones have gain-ranging A/D converters, for example, but they offer no wider dynamic range than comparable analog microphones that are connected to external preamps and converters--a comparison that can be made by anyone, since Neumann sells several of their models in both digital and analog versions. And for the record, those mikes all have digitally remote-controlled gain settings.
--Someone further up the thread surmised that 32-bit float must be a superior format because why else does so much audio software use it internally. The answer is that recent Intel CPUs can process four 32-bit floats in parallel for many mathematical operations, which makes DSP functions in the software much more efficient than linear PCM allows, if your CPU supports the needed instruction set.
This may not be so obvious, but the real issue is the level and behavior of the recorder's internal noise floor, so please keep that issue in mind from now on, OK?
The two techniques that we're talking about can be understood as the same thing in different guises. They both involve "tracking" a signal in real time (i.e. at the sampling frequency) and then, based on its voltage at a given instant, assigning it to a category of bigness or smallness, with a further number of bits indicating where the particular sample value fits within that category. The dual-ranging-converter arrangement is the more obvious of the two in how it works. 32-bit floating point (at least the IEEE 754 flavor that I assume they're using) is the same thing, just done with 256 overlapping levels instead of two.
Both technologies cause the noise floor of the recording channel to rise or fall in response to instantaneous signal levels. When the signal level goes up in magnitude (i.e. absolute value) and you move into a/the higher range, the noise floor of the channel rises along with it. When the signal comes back down, so does the channel's noise. As long as that noise floor is so low that you can't hear it (always, 100%, money-back guaranteed under all circumstances), then the fact that it's shifting up and down will be of no audible consequence.
But that's the big "if" right there. If the noise floor is ever audible--if any possible type of signal, or combination of settings and signals, can coax it out of hiding--then it will be heard to "pump" or "breathe" along with the momentary signal levels. That will make the program material sound gritty or dirty or some such unwanted thing (depending on implementation details such as pre-emphasis/de-emphasis).
It's an effect that used to be called "modulation noise" back in the era of analog tape, and it's one of those things that once you've noticed it, you can never un-notice it again. On wide-dynamic-range program material where the levels change quickly by large amounts, noise pumping "calls attention to itself" and is far more offensive to the ear than a steady, low level of broadband noise would have been. And there's no real way to get rid of those artifacts once you have them on your recording, except to cover them with high levels of steady noise, which is obviously undesirable as well.
So the only hope for this recorder to sound good is if its internal noise floor is so low that it is never, ever heard, even "out of the corners of one's ears." It's possible, but by no means guaranteed. The gain manipulation in the A/D converters and the floating point encoder means that noise caused by those elements of the system will constantly shift up and down. The floating-point encoding system actually doesn't worry me unless it's implemented in an almost unimaginably, bone-headedly stupid way that surely someone would have caught and fixed by now (except, the people at dbx back in the day claimed that they really didn't hear the noise breathing of their noise reduction systems, which was pretty horrible at times). And there are ways to get dual A/Ds to play nice together.
But it all comes down to the actual implementation. If that is as good as it possibly can be, then we'll have a recorder with no gain control, that by definition can't have a wider dynamic range than the best previous recorder that has a gain control. So before you fall too far in love with this design concept, I suggest that you imagine epoxying the gain control on your best existing recorder to a setting that you know will never allow overload. Do you think that it would always make recordings that are as quiet as you could have made if you'd set your levels specifically for each occasion? I don't think so. And in that case you shouldn't expect more from this recorder.
--best regards