That n-Body simulation is a bad case for Python today in that you loop over the differential equation solver in Python. (It's like the very branchy semantic web and old AI stuff that I do for fun... I am migrating a lot of that to Java, PyPy helps a great deal but Java does a lot better.)

If you are doing heavy matrix math, numpy runs at FORTRAN speed, tools like scikit-learn and Tensorflow also get high performance by doing the heavy lifting outside Python.