Depends on how you understand "goes down". If talking about the network as a whole: True. If talking about the likeliness of communication between two random actors being temporarily interrupted without notice or workaround by the system itself: Unsure.
Agree, but that’s not what fragile means. Maybe unreliable? But how many times does your scenario play out? How unreliable does something have to be to be officially “unreliable”? I’ve only had post not propagate because of clients, not the protocol.
I'm not sure whether this is really the point - or rather the fact that this kind of (obvious) unreliability, including measuring or benchmarking it, hasn't apparently been considered on protocol level. As a participant, an unreliable communication system is (strictly speaking) even worse than one that visibly fails ever then and now...