Someone mentioned the brittleness of Chord.

Thank you for making the only insightful comment I've seen in all of the web discussion. You are quite correct that Chord is generally considered to be brittle, and so might not work as advertised in the paper.

Claims of Chord's brittleness are entirely theoretical at this point, but certainly not to be ignored. Real world testing is needed. There are other distributed hashtable implementations which could be dropped in place instead of Chord. Kademlia is quite the rage these days in the p2p-hacker community. Also Pastry and friends. These all need further testing as well to know how they work in high churn environments. I also have some hacks to make Chord more churn-resistant. It's an exciting area of research.