what do you think?
Try expanding the idea to a more formal or longer paper with diagrams.
But if multiple hosts can use the same ip how the connection can be traced? Only the ISP have the information, the receiver don't know anything a part from the content of the packet.
Your suggested adversaries can and will run a 'receiver' to get the contents. They are in the 'receiver's ISP and use netflow or other means to trace back the spoofed packets. It's basic network administration.
message could be destined for anyone using public key encryption: if A sends a message to B and B can't decrypt it, it wasn't intended for B, so it gets forwarded to other nodes in the network.
Unless sender A is aware that final delivery was made to Z (or knows the net is reliable), broadcast models will fail to deliver reliably due to being clogged out of the links or aging. To be reliable, broadcast needs control, knowledge, or maybe unavailably large time/space.
Traffic analysis is defeated by layering encryption and constantly sending lots of flak: nonsense messages. If you can maintain the throughput at each node as a constant and make one message look different between entering and exiting a node, I believe it would be theoretically impossible to conduct traffic analysis.
Packet sizes must also be the same throughout the network. And must be detectably immutable at each layer of link and onion-ish path encapsulation, or be dropped. ~1500 mtu minus layers = data capacity. Related... redundancy to the destination could serve as chaff. Yet with underlying privacy, unclear on the need for redundancy (as chaff), unless the redundancy solves full path (or node) reliability issues. Otherwise 'control as chaff' seems more valuable.