"Obviously it would have to come with some lower level semantics, like memory management."

Why? Garbage collection is nowhere near the top reason high level languages are slow, and there are GC languages that are fast.

"By high level, I meant higher level than c++. it will always be lower level than a language like python"

So, Java or C#, then? JIT languages that can be as fast as C/C++ in some circumstances, and faster in others (and slower in still others). And you don't even have to manage your own memory, so it's seemingly higher level than you're asking for.

I think GC is fine for high performance computing. Might not be fine for real time systems or games, but that's not what this article is about. It's about high performance scientific computing in Python. Having a GC sweep every now and then, even at unpredictable times, isn't going to cause things to come to a crashing halt or ruin the performance of a batch run.

PyPy might be something you'd enjoy reading about. It is an alternative implementation of Python (mostly written in Python, and using a JIT), with the goal of being fast like C. This may be the closest thing to a CoffeeScript like tool where the goal is to make a very high level language faster through translation or "compilation" or whatever.