I haven't taken a look at the actual programs, but given the nature of the benchmarks the result can probably be interpreted as 'numpy is still quite fast' also?
However I think the basis of these benchmarks is really to pit languages against each other and not it's frameworks/libraries.