Why does Bitcoin use a gossip protocol rather than a DHT for peer discovery? I believe that kademlia XOR distance helps shape the network topology so that peers are more evenly disturbed.
DHTs are built so that nodes keep only a small portion of other nodes addresses in their routing table (the closest ones). This means that when you send a message to a DHT to put("foo", "bar"), only specific peers are going to see it (in fact as little as possible) and even fewer are going to store it. So if you want to do "peer lookup" using DHTs, it wouldn't work because nodes aren't programmed to send messages through a path to all nodes.
Also I think you may be confused with BitTorrent's DHT. They store IP addressed of peers downloading torrents as values of keys in the DHT. So in that sense, they do use the DHT to "lookup peers" downloading a specific torrent. BitTorrent is a different system than Bitcoin, which is not a file-transfer protocol, and therefore I don't see how a DHT can be useful to Bitcoin's design.