Now that R2013b has finally incorporated Bruno's efficient implementation (FEX 24499) into the built-in KRON function, I expect the balance of performance and memory to shift in favor of KRON. Of course, expand will continue to be better for anyone still using R2013a or earlier.
For additional alternative implementations, see http://www.mathworks.com/matlabcentral/newsreader/view_thread/326075
just like jos, i also felt at first that this - admittedly sleek engine - was more or less a KRON clone; but then your timing convinced me that there was clearly more to it...
however, i think you should at least mention KRON in a