I am wondering the same. I have a lot of SciPy and NetworkX based simulation code I have been meaning to plug into web interfaces. This could be extremely useful for academic research.
Not that I expected otherwise -- there are plenty of people (like, say, my officemate) who would pay a few bucks to run a few million simulations on Google's infrastructure. (I would too, but it'd have to be a lot of nodes, since I have my own farm of client machines due to the way I arrange consulting)
I hear they are actually CPU starved these days, which quite literally boggles my mind... if Google can run out of cycles, anyone can. But that informs the current deployment and limitations if it is true.