So what is the verdict? Is Numba indeed consistently outperforming Julia even after the code optimization?
The verdict is that the improved Julia version is always faster than numba. For the original choice of parameters the speeds are very close, Julia about 7% faster. For other combinations, Julia can be faster by several factors. Julia is better for smaller xs, presumably because numba has some overhead that is more noticeable for shorter loops.
BTW, on hacker news they call this Ā“nerd snipingĀ“ Julia;)
Perhaps you can adapt the title accordingly?