Much of the problem lies in set-ups. Distributing the required code across corporate lines is a tricky business, when you have a commercial end in mind.
Another problem lies in Moore's Law. As individual computers and small networks become more powerful, and even desktop PCs start applying the lessons of parallel processing which grids taught, the need for grids decreases.
Thus they remain, on the whole, an academic exercise, not a market phenomenon.
Demonstrations like this one between a European and Australian grid might cause us to reconsider. The two grids, half a world apart, cooperated in simulations of HIV drugs.
One thing I've learned writing about science, especially as practiced at my old alma mater, is that the intersection between big computing and big breakthroughs is getting to be like an Interstate junction.
In my view every medical research facility should have a grid, and the ability to connect this grid with others, in order to plow through these kinds of jobs and create breakthroughs.
Such grids already exist. All they need do is organize and market themselves properly. Things like this DEISA-gridaustralia hookup need to become routine.
UPDATE: It seems I've hit upon a Google business strategy here.