I don't mind about knowing limitations. I'm saying that these optimizations usually are very hard tradeoffs and opposite side of it β code readability, speed of developer work, ability to maintain it working. I tried Cython and PyPy. Both are really good if your project is started with them, but if you decide to migrate to them in order to increase performance, it's like rewrite project to another language. Also both have a lot of limitations and cpython gives you still a lot more flexibility in decision making (choose frameworks, libraries and approaches how to solve certain problem)
Learning the tool you're working with means you know patterns to write generally more efficient code.
Even if you're going to use numpy/cython/cffi for faster submodules, writing faster code in general is a good thing.