Razor, you're right: all audio data ultimately ends up being 16- or 24-bit integer data. Even if you play it back using software that supports 32- or 64-bit floating-point files, it will have to convert it before sending to the audio interface. The only reason you'd export floating-point data is if you plan to subsequently import it into another DAW or audio editor for further processing.
IMO there is absolutely no reason to ever create 64-bit audio files, unless you're sending them off to someone who has specifically asked for 64-bit files.