If you plan on building any p2p network, Iroh should be the first thing to try, and you most likely won't try anything else, because it just works.
Caveat is; make sure you actually need p2p, because Iroh can't magically solve all the other challenges inherit at leaderless distributed systems.
Yes I think they did very well for p2p applications.
Many (most?) applications (inlcuding nostr) are architected to use servers for good reason. I define servers as machines that are expected to be up and available nearly all of the time. You could also say that servers must accept incoming connections, but that isn't strictly neccessary. With Iroh you can run servers behind CGNAT and have them accept incoming connections via Iroh. So servers no longer need to be on the bare Internet, or NAT forwarded... but they still must be almost always available. And this aspect of availability makes a huge difference when architecting any kind of application or protocol.
Make it easy for users to setup a geographically distributed cluster (servers hosted at 3 or 5 friends and family homes). This could help make it easier to get high availability?
Yes servers are just nodes with a human that can be bothered with maintenance, because so far no one invented a server that doesn't need baby setting.
Http tunneling makes more sense than Iroh for servers behind NAT, because this way the client doesn't need special software.
Really the main reason to use Iroh is if the node is run by someone who isn't technical, or for mobile phones where there is simply no way to open a port or have a static IP.
