Anyway, my thought on why numpy seems more widespread is simple: efficiency.
Even Matlab code gets rewritten if it needs to be used for something intense, and Matlab has many more optimizations than Octave to get away with sloppy coding. Python+numpy is pretty easy even for a non-programmer to write (at least compared to C++ or whatever a DSP wants you to use) and even sloppily done is generally fast, plus it has a great and large community. But both Matlab and Octave are great (at least if you use octave-forge) for prototyping and doing a lot of domain specific things python/numpy/scipy don't support out of the box. Matlab especially has a lot of neat GUI helper tools. Meanwhile R is over in its own world where if you're doing anything in that domain it's a questionable choice not to take advantage.