Your code has to be explicitly parallel, it won't be parallelized automatically. As you are running it, you are just running copies of the same thing on the compute nodes.
Examine the octave examples, you'll see how the parallelization is done. The work that is done depends on the mpi rank of the worker the code runs on.
There is some parallel mp3 code available, see
http://www.osl.iu.edu/~jsquyres/bladeenc/I don't know about the quality of the resulting mp3s, but it's a place to get started.