Sounds a lot like the sort of thing that Apache Kafka does - just passing around messages. The mempool/p2p network of blockchains basically does the same thing, it doesn't keep anything permanently either.
The tricky part with it is how to pay the providers.
I would say just run a bunch of UIs on top of it and have it run like e.g. Mastodon. You would basically use it like Twitter anyway. You can set up anyone to accept or reject what you want. I was thinking of slightly reworking the original hashcash stamp format with hashcash over a timestamp to have a rough estimate for how long a message should remain based on investment in preparing in (but no guarantees, you still want to be able to be selective on what your node considers spam for example...)
Talking about providers, I think hashcash on its own can provide an interesting pay-for-bandwith model without any money involved, whereas for paying actual infrastructure would be more Bittorrent like: we run it because we want it.
'we run it because we want it' does not become competition for anything. Bittorrent survives because the file sharers were able to weather the political siege whereas the subject ran out of political currency. You wouldn't call bittorrent a service, since it's purely enthusiasts. All bittorrent had to do was separate the peers from the trackers and voila, legal teflon. Not that they haven't and still aren't hurling DMCA takedowns left right and center. It's just fishing trips anyway.
You do realise that hashcash is the very basis of the Bitcoin Proof of Work consensus, right? Well, unless it's worth nothing then if it's worth nothing it's not gonna be expensive to break yada yada yada.
No, the future of internet forums is looking at you right now, it just needs to have some interconnection protocols added, and clients that indiscriminately allow access to any and all forum systems with as little complication as possible. It won't be one network, it will be many. The biggest obstacle is bringing enough of the users of specialist areas into a single pool of connected networks.
We see in these times with the graphene social chains, that this is partly a matter of politics, and partly because the protocol is only somewhat decentralised - decentralised up to the point it costs more than 3 seconds a block. Cutting latency from decentralised networks requires culling the number of nodes that must validate the new transactions. Thus 19 witnesses, and not 9000+ miners like with bitcoin.
The very same tech used to run Facebork and Twatter is also used by blockchains, in their p2p network between validator (miner, witness) nodes. They can have so much bigger userbase because they are more focused on network latency and make no promises about immutability or ownership of the data users put on the network.
It's my opinion that decentralised networks will replace these centralised pestholes once people start to grasp the idea of data that does not need to be kept forever. There is already protocols to pay people for delivering data around the internet (maidsafe, storj, tachyon) and those protocols combined with kafka style partition resistant message relay network systems, with a way to be paid for relaying...
It's a slow process. But the big players already depend on many of the techniques used in blockchains, blockchains just have a few stupid ideals that need to be understood as having a specific place relating to the fight to eliminate the cartelised banking system, and fighting facebook and twitter will not make any progress until the idea of a wibbly wobbly consensus state is accepted. Cos that's how they do it, except you can't install an app on a server and join their network, and get paid for helping deliver content to users.
It's coming, as surely as the sun will rise tomorrow as the earth spins around.