Well then I would say that you’re stuck with essentially the same problem no matter what you do since Pro Tools also uses floating point processing internally. The only difference then would as you point out be the choice in dither.
But let’s say you’re stuck using this method, I have to ask how low the vocal signal level is. For the dither to be heard it’ll need to be close enough to the actual signal. I would imagine then that raising the outgoing level of the channel that you have processing on before it hits the track you record onto would possibly solve this problem.
Something else to consider would be to import 24-bit files but set your project to work with 32-bit float files. So any newly recorded files would be float, and then you can choose your own dither that you prefer. After all, any processing you do in 24-bit would eventually either cause quantization distortion or get dithered, no?
Lastly, I seem to recall several threads about this on Gearslutz, but I can’t remember what the final discoveries were. Basically a 32-bit float file has the same ‘resolution’ as a 24-bit fixed file, because the ‘other’ bits are not used for resolution but for scaling. So in a sense you’re not dithering in float, and there’s the case to be made for not doing it going to 24-bit fixed either (if I remember correctly). So I’d definitely say you should search on Gearslutz or elsewhere because this is exactly the type of stuff that DAW makers won’t discuss publicly for some bizarre reason.
PS: Actually, one more thing… Is it possible that the first instance where you ran into this ‘issue’ was simply a case of the plugin not acting the same in the two DAWs? I would think that it would be hypothetically possible. And if that was the case then what was audible could have been that problem rather than the eventual dithering going from float to fixed… If you do this for a living I would highly recommend looking into that as a possibility, even if it just serves to eliminate that as a problem. I would guess you should be able to for example import a file and set AT to a specific setting(s) in both apps, and then render out the same way using the same process for dithering. If a polarity inverse and sum results in only dither noise then it would seem AT isn’t an issue, but if you hear some weird voice artifacts then perhaps the plugin acts differently in different DAWs. And perhaps test with some automation as well…
just some food for thought…