To be clear, Keenan, Julia's randn() function usually creates Complex floats, which is a Julia datatype that handles really large floats. Of course, Python just ignores the complex part and considers them floats. I performed the test as well, and got about 4 seconds, so Julia still offered a speed improvement.

https://github.com/emmettgb/Emmetts-DS-NoteBooks/blob/master/Python3/numpyvspythonspeedtest.ipynb

Unfortunately, I cannot upload the data to Github as it is too large, but for a test that makes a little more sense you could always time Julia performing the operations on the data?