Numpy probably.

> Naturally far more efficient

Most data science languages are interpreted with a few higher-level routines. Same with J, for instance - it has adequate performance as well.