S/Kademlia does not solve this problem; it simply slows down the rate at which an adversary can attack the system by a small amount (i.e. by making node ID creation more expensive and increasing a key/value pair's number of replicas).

There are several DHT papers that talk about bootstrapping DHTs off of social networks. They all fail to solve the Sybil problem in the same way: an adversary simply attacks the social network by pretending to be many people.