I'm afraid you only take into account private use.
On-endpoint filtering is not enough on a large scale. In a network with expected high level of security I don't trust the endpoint and censoring them is a feature, there is no moral or legal expectation of privacy, but data integrity would be nice.
If you own all of the endpoints, then it is indeed morally okay for you to do network-level filtering. The problem is that if we design network protocols so it's possible for you to do that, then it will also be possible for people to do network-level filtering that affects endpoints they don't own.
