I agree with you, the cache should not contain repetitions, also because otherwise the algorithm fails (as showed before).
Also, I agree that a peer should not select itself, since it would be a null cycle: the peer exchanges the cache with itself, so each entry is replaced with the same version of itself, and so on.
I do not agree that the cache should not contain a copy of the data of the peer owner of the cache: not because it's wrong (in fact it's not only useless but disadvantageous, since it wastes a cache position and reduces the degree of the node by 1), but because, unlike the cache repetition problem which it's unclear but can be deduced from the text, the fact that a peer contains an entry of itself is explicitly stated in the paper:
Each correspondent executes the following five steps once every ∆T time units.
- Request a fresh news item from the local agent by calling getNews(). Create a new cache entry and insert it into the cache.
- [...]