I'm a senior majoring in physics and doing research on the the Epoch of Reionization with a radio cosmology group. Most people, at least in the research group, are on mac's as am I. This, I suspect, is mostly due to them being unix boxes with a nice GUI. I'm not sure what software people studying GR normally use but I end up using a lot of Mathematica, IDL, and Python. My little macbook air seems to work well enough, I can do development, run some stuffy locally for quick tests, and spin all the big stuff off onto a cluster. I have noticed that doing some fun integrals in Mathematica involving QM can easily spike my CPU's for a bit but the convenience is worth it. Something that is easy to take to lab meetings to show people your pretty data is fairly important.
In my experience most scientific software, such as those listed above, seems to be available on Mac/Windows/Linux and work about the same. One downside to running Windows though would be that if you are going to be interacting much with a cluster a Linux/Mac system will allow you to more accurately test things locally such as bash/zsh/fish scripts that fire off your analysis program on a cluster or reorganize large amounts of data. A fairly easy workaround would probably be to just install Cygwin on Windows but I have little experience with that.