I just read an interesting paper recently released by the team at Microsoft on the topic of centralized Mega-data centers vs. geo-distributed micro centers. The authors argue the benefits of the distributed approach and it's a glimpse into what the future likely holds for "Embarrassingly Distributed" cloud services. I think one of the more interesting concepts is the notion of using smaller facilities to sidestep some of the costs of power transformation and network routing. By dropping many small data centers onto the open-market's mesh of existing infrastructure some of the costs of the mega center are potentially avoided (primarily initial infrastructure cost). I'm not sure all the costs of operating in a distributed mode are well explored in the paper or can really overcome the operating economies of the mega center, but they describe other clear benefits to geo-distribution. The case they make around Hotmail, as relates to eliminating spam at the incoming edge, is very interesting.
Worth a read if you're following what the big cloud service providers are thinking about these days as they deal with the phenomenal growth of their platforms.