If censorship of some kind is implemented, it’s implemented. If you dislike it, better to ensure it is not implemented. Messing around with deduplication misses the point: don’t allow censorship or find efficient ways (eg increased replication) to nullify it.
I’m not sure of the maths, but even for large files I suspect it would require a significant proportion of all nodes to opt in to censorship of a particular file to ensure its deletion. Which in a way is a form of concensus.
Also, the network won’t be the sole holder of most data so it’s not necessarily lost. If it is important it’s likely to be recoverable somewhere.
I’m not arguing for censorship here, like others I think that any filtering should be application layer and under the control of individuals. My point is that we don’t need to be as concerned as some are because as things stand it is not likely to be implemented and should not be used to divert the team or change the implementation.
The point of this debate is to anticipate legal and regulatory moves which might undermine the Safe fundamentals and counter them.