Thanks for the reply - that has been the assumption I've made - 32-bit float is 32-bit float unless the app fails to dither at the final output, or a plugin is garbage, or you track/monitor with an average-to-lousy front end. I'm running an RME Digiface to my Paris rig (using Paris 24-bit converters for now, and adding an RME ADI-8 DS or other converters in the future). Clocking is of course another issue, but I'm not sure how Logic's internal audio playback/recording clocking is (i.e. if running alone, hardware internally clocked).
I have a UAD-1, but haven't tried it yet (still building the PC). It is a VST/DSP card - shows up as regular VST plugins in a compatible app (currently PC-only, Mac coming soon). Word is that it sounds fantastic. Some say better than the TDM and native equivalents, and the claim is the emulations of LA2A, 1176, etc are quite close to the hardware originals. The reverb is supposed to be superb (better than native) but I haven't heard any specific comparisons to hardware yet.
I am thinking POW-r dithering in Logic 5 may answer the question of what happens to the 32-bit float mix sum when output to a 24-bit file (truncated currently I would guess). This could have a significant impact on sound quality of mixes in Logic. I don't know if Emagic is implementing POW-r in other portions of the processing chain. I've posted to the LUG to try to find out what processing is done along the path, but no response so far (it may be 32-bit float from start to finish, then truncated to 24-bit at the master output/bounce stage, but it is hard to say for sure).
Regards,
Dedric