I also suspect you're overestimating the cost of running a HAF server because you're only looking at "full" HAF servers that capture all blockchain data
For sure. When we wanted to develop the polls protocol using a regular HAF app, the limiting factor wasn't HAF (which we could fill up with only the data we need) but hived. We couldn't run a tiny HAF without hived, so we started looking for some other way. So this is how the idea of the standalone HAF app came about and the reason behind it.
So what I ultimately see happening is people running lightweight HAF servers (i.e. ones that filter operations/transactions they are not interested in) and some amount of replication of the tables of these databases on subscribing databases.
Sounds like a viable way to do it. Just thinking through it for the polls protocol. If we used publish/subscribe, we'd need hived + HAF filled with polls custom_jsons and account data. And then we can subscribe to those particular tables. But we end up still needing hived. Is there any way you can see us avoiding hived in this kind of setup?