Should I develop metaheuristics like GA, ACO, PSO etc. using Matlab, or MEX and C++, or purely C++, Java etc.

I have been working around solving different optimization problems, mostly NP hard combinatorial optimization problems, such as TSP, job shop scheduling, VRP etc. I'd like to develop metaheuristic algorithms. However, in many cases, loops cannot be avoided because iteration is needed. For some complex problems (such as flexible job shop scheduling), even in decoding of chromosomes in GA I have to use loops. Will it be more efficient if I write whese codes in C++ and use MEX to call them in Matlab? Or is it a better choice to switch to C++ or Java etc. for the entire project. I am familiar with Matlab and I can easily write Matlab versions of programs, and do vectorization as far as I could. It will take me some time to switch to C++ or Java etc, however they are definately more efficient with loops. Does it worth?