This was an old thread that came to life I see.
I don't think any mastering engineer should have a problem with 24-bit audio files, and if you've clipped your stuff in what you deliver, well, they should just send it back. 32-bit is kind of like the 192kHz thing, but not quite...
Internally, the value of processing floating point numbers is simple, every [Intel] processor since the 486 has included a floating point processor, or unit (FPU). This allows that [floating point] arithmetic, of which there is a bunch, to all be done on a separate processor from the one doing all the other stuff, like running drivers for devices, file i/o, moving blocks of data around, etc. (If it had to work all the time with integer data, it would be converting it back and forth to floating point too, since all of the algorithms for transforming data work with floating point numbers.)
Externally, it means nothing, unless you're in the habit of bouncing/transferring audio at 0dBFS+. (IMHO)