Well really, C++ is rapidly replacing Fortran for scientific computing, but it ain't all that much better for it. I'm excited about the idea that this replaces MPI.
Physicists at my school/uni (in France) use Python with appropriate libs, but I don't know if it's the general case here, since I know physicists from near university use C++ (badly from what I saw, but that's an other story).
I'm a physicist, and I use Python (with scipy/numpy when appropriate) when I can get away with it (ie for small to medium-sized calculations).
For big calculations, though, where what you can do is limited by performance, there's no getting away from C++ or Fortran. Python has a huge performance overhead (I have learned this the hard way).