The Sobol.jl package was originally a wrapper around a fairly optimized C implementation of the Sobol sequence (from NLopt), but Ken-B just ported my original C code to pure Julia. The performance of the Julia implementation is essentially indistinguishable from the C speed.
It's not too surprising since the algorithm is mainly bit twiddling, but it's always nice to see C performance on "real" problems compared to optimized C implementations.
