Jim Gray of Microsoft Research has written a paper on the economics of distributed computing and the findings, although perhaps obvious, are thought provoking. His conclusion is that the costs associated with computation are so significantly cheaper than the costs of bandwidth that distributed systems need to leverage computational power and perform most of the computing operations near the data. The trend up until very recently was to distribute as much of the computation power as possible to promote scalability, leveraging networks to do so. With more applications being distributed across public networks, the economics of this approach may not make sense.
I highly recommend that you read this. Also, if you haven't seen Jim Gray speak before I highly recommend that you try to catch him at a Microsoft event or in an online seminar.