The general idea is that people own their data. So if you want to make a post on a forum you will pay for the PUT of your own post and then give the forum access to read the post. It would be technically possible for someone else to pay for PUTting your data, but then they decide permissions etc.
As for who pays for the traffic, my understanding is that it has to be paid for by everyone who PUTs. I don’t know enough about economics to know exactly what the chain of “events” that causes the prices of PUTs to increase when a ton of people want to download stuff from the network. But I do know that payment for PUTs is the only “source of income” for the network, so it’s the PUTters who will inevitably pay for running the network.
Yeah you only pay once, but as long as some people keep PUTting stuff, farmers are compensated for the cost of heavy traffic for instance by an increased PUT price, at least when measured in fiat.
Nodes will not be hosting “sites”. A node might be hosting a part of an image of a site, or the text of a page on a site for instance. When something is uploaded to the network it’s spread out across it, so no single node will be hosting “a site”, the network as a whole hosts it, but no individual node does. This distributes stress across the network, so there won’t be some nodes that have to deal with much more work than others.
There have been many changes to how the network works over the years, so the following might not apply to how it’s working now, maybe someone who knows more can clarify if that’s the case:
A planned feature that will help when some specific content is requested a lot is caching. This feature will make it so as chunks hop across nodes on the network they are cached by the nodes that transmit it, if a request for a chunk arrives at a node that has that chunk in its cache, it will send its copy instead of asking the “original holders”. So chunks that are popular will put less strain on the network per request the more popular they are.
The reason I think this might not apply to the most recent version of the network is that I think the network structure was flattened recently. So each section can communicate directly with every other section. So when a request comes in it could send that request on directly to where it needs to go, and then that section can respond directly to the client who asks for it? Not sure how this works now.