You are imagining things. In a properly executed blind test you will NOT be able to hear any difference. You are “hearing” a difference because it is a bigger number of bits and you want to hear it due to the psychological bias you have. Also, if you are not doing any offline processing there is ZERO difference between recording 24 bit fixed and 32 bit float.
The only way you will see a difference is to intentionally force errors using offline processing as Soul-Burn described, but if you have anywhere near remotely decent gain-staging habits, 24 bit fixed is more than adequate resolution and the mixer is always floating-point no matter what file type you use.
Last, with modern computers I do not believe you will see any “less processing” load on the computer by recording 32 bit float files. The files are bigger too, and if you record much this can be a penalty (despite what others may claim). I record many tracks every day and back them up after every session. File size is still an issue for me even with today’s large inexpensive drives.