> If you also prioritise files that have the least replicas across the entire network, and allow servers to re-distribute the content they are replicating, it would surely become quite hard for a file to be lost forever.

How would this stand up to an attack of uploading random data in large quantities?