Since numba (https://numba.pydata.org/) and Julia have similar aims, it would be nice to either have the python benchmark converted to use numba altogether (I guess this is similar to the choice of using LuaJIT) or a separate numba benchmark.