If the question is, “Should I expect any performance difference for the same CUDA application running on identical hardware, but booted to different operating systems”, the answer is “Possibly yes”.
(1) The OS, in particular system libraries, may impact the performance of the host portion of your CUDA application.
(2) The driver architecture imposed by the OS may affect the performance of the CUDA driver. In particular, performance may be negatively affected by the default Windows WDDM driver model. On Windows, if possible, you would want to use the TCC driver model (this is only supported by selected GPUs).
It’s impossible to say what the largest performance difference you could observe for the same app across multiple OSes on the same hardware. Personally, I have seen differences up to about 10%.