That doesn't really make any difference, AFAIK. A floating point engine obviously has some advantages, but if you lower all the levels of all tracks, process them at that lower level, then push them back up, that's exactly what bad gain staging is.
If a floating point engine made gain staging a non-issue, you could turn down the sensitivity on your converters and store everything at like -50dB (analog input chain gain staging would be done as normal) and do the whole mix and push it up 50dB at the end and have it sound just as good as if you'd stored it at more normal levels. I don't think that's the case, right? All the processing introduces noise, and the more bits they have to work with, the more the signal stays up above that noise.
Seems to make sense to me anyway.