4-byte IEEE float I'm pretty sure will give you lower quality loss in the long run. By that I mean when you effect the track and process it a great deal. There's more bits for rounding error by the plugins.
For a few tracks I don't think there's too much difference between 16 vs 24bit but I would still always use 32 bit (24bit) if your soundcard and system support it.