@kristoffer.carlsson agreed; that article about pyston was the first thing that came to my mind.
Another piece of the “why Julia” puzzle is the fact that Julia does not treat a specific set of built-in types differently. That is, I can make my own type or immutable that can do anything that Julia’s built-in types can do, and my type can be just as fast as the one provided with Julia (if I were as skilled at numerical computing as the devs, which I am not). This isn’t just a nice feature; it’s essential to making things like automatic differentiation and other numeric types fast and easy to use. It’s also why we have really nice abstractions like the idea that an image is not a matrix of doubles or bytes but instead a matrix of colors. That works because Julia lets us define a whole family of color-related datatypes which provide useful behavior and abstraction at little or no run-time cost. That’s been the biggest benefit of Julia, in my experience.
Oh, and also JuMP. JuMP is great.