Basically, the reason MATLAB/Numpy/R folks harp so much about vectorizing code is because their interpreters are slow, and within a vectorized function they can punt to some C/C++/Cython implementation. But Julia loops compile down to almost ideal assembly; they are remarkably fast. In fact, verbose loops acting in-place almost always beat the vectorized alternatives because they don't need to allocate space for the result.

The wonderful thing about Julia is that you can code it vectorized first, and if that's not fast enough (or if it uses too much memory), you can easily rewrite it in the same language.