Oh yeah, a different cluster is the obvious solution (or like gnuite said use ext2 and write a proggy to ln all of the files with the same md5 to a common file). I was thinking of people possibly wanting to shuffle data of commonly used cities that do not understand this issue.

I've been playing with the idea of server. Like said, there is alot of redundant data. I could further subdivide the data and really clean house. Just rebuild the zoom level 0 and server up the rest in the 256x256 blocks.