Looping within M file is kind of a nightmare years ago, as there was almost no optimization for the loops, however, nowadays, the latest version of Matlab has already made loops faster than before, it seems there is no need with manual vectorization.
It is worth to point out, writing loops within command window environment won’t get a lot of fun, since JIT will not optimize loops there, try to write a script and run, this will automatically call JIT compiler to optimize codes.
Back to several years ago, there are lots of self-tuned MEX codes using BLAS/LAPACK to overcome the setback of Matlab’s for loop pitfall.
I recompiled mmx and mtimesx and tested them with our 2014b for loop, the result seems that for loop no longer stays the worst position.
However, since for loop on matrix multiplications are using multiple threads, it is still faster when making comparison on usual matrix computations. While when computing operations with ND arrays or ND matrix, mmx and mtimesx will give a few multiple of boosts.