There are always tradeoffs. DHT resolution may never be as fast centralized solutions. And as far as I know there's still no general solution to certain attacks like sybil. An intermediate solution that works with today's technology is for people to self-host "partially centralized" technologies, a la federation, but the UX is currently prohibitive. I don't want to manage 5+ instances for all the services I'm interested in and I also don't want to pay $5/mo to have someone else manage all of them, and still be stuck coordinating all those accounts. I'd love to see something like sandstorm.io as a service become popular for federated tech.
DHT perf, sybil resilience, and metadata exposure are weak points in p2p. There are ways to mitigate its necessity: SSB removes the DHT entirely and gossips data through a topology which users configure (on network or via manual setup). At this stage I think you want a hybrid so that you can reliably find data via the DHT, but then move away from the DHT when possible. Once you have reliable supernodes, its feasible that you'd check with them first over an existing connection prior to hitting the DHT.
I don't see how you can suitably solve the ops of federation because you require consumers to administer servers. The hope with P2P is that you can make ops relatively simpler; you remove servers, run the business logic client-side, and build against a distributed CDN. The supernodes will require administration, but, because they're dumb/unopinionated services, they are able to support a variety of applications.
I think the human cost aspects of federation could work like subreddits in terms of the amount of work necessary to keep things running relatively smoothly. The tooling likely needs to get way better though.
