Momentum, random names and (possibly) higher latency. IIRC IPFS is particularly bad latency-wise but I'm not sure how much of that is name lookup vs file transfer, and that could be implementation specific. Name lookups are also very cacheable.
I think what's missing is that squatters and other bad actors are going to attack the distributed name system, too. It should be somehow resilient, and ideally resistant, against deliberate misuse by powerful parties. For instance, DoS attacks that pollute the namespace and could make the distributed naming service too slow or resource-intensive will necessarily be mounted.
This is one place where a significant proof of work, along the lines of Namecoin or handshake.org, would make sense. (Another place is password hashing, for example.)
> IPFS is particularly bad latency-wise I'm not sure how much of that is name lookup
All of that is due to their mistake of trying to use a sessionful protocol for their DHT.
Bittorrent got this right -- sessionless DHT -- which is why IPFS remains a rounding error compared to bittorrent, and will remain so until they adopt a sessionless DHT.