Main problem appears to be construct validity. If they're building a computationally vast model based on impoverished data (i.e., a pre-selected subset of current morphological/functional knowledge), nobody will be able to determine whether its electrical behavior tells you anything at all about a real brain.
More to the point: If they want to get this thing running within ten years, they'll have to simplify at each step without knowing the potential repercussions of each simplification. I thus find it disingenuous to call their project a "model of the brain." It's an implementation of a very specific reduction of neurobiological state of the art. Sure, that can be useful, but only within strictly confined limits. You won't be able to do drug tests on this thing -- so much is certain.
From what I've read about Blue Brain it appears that validation is a central theme of their work. They are actively validating the small piece that they have already built (the neocortical column).
More to the point: If they want to get this thing running within ten years, they'll have to simplify at each step without knowing the potential repercussions of each simplification. I thus find it disingenuous to call their project a "model of the brain." It's an implementation of a very specific reduction of neurobiological state of the art. Sure, that can be useful, but only within strictly confined limits. You won't be able to do drug tests on this thing -- so much is certain.