The problem is that you may have to stray outside that vectorized bubble, and at that point you enter into a two-language situation if you are using R (e.g. RCpp) or Python (Cython?) - which was one of the motivations for the whole Julia language, to avoid needing to do that.