Well for one thing : millions of lines of code use numpy, so porting Numpy to pypy would be a great step forward for the porting of many other (scientific and other) librairies to Pypy.
Re-implementing all Numpy in pypy may sound like a chore, and it is, but the alternative in just insane : re-implementing hundreds or librairies to use XXX instead.
And by the way I don't know theano really well, but I know it isn't a substitute for Numpy. they are different projects, with different features.