Yes.
In theory, miners can prune state that is already migrated to the current epoch from the previous. Should they choose to do so, there would be no data duplication. While the state root for the previous epoch cannot change, they may need to generate a proof for state that is the sibling of a pruned node. Luckily, they only need the hash of the pruned branch to generate such a proof, not the data itself.
That being said, pruning state is actually a hard problem given current client database design so an alternative database design (maybe only for the previous epoch branch) may be necessary to enable better pruning.
It would be very valuable to get an idea of how much state is active vs inactive though, so we can identify whether this is something we should think harder about or ignore.
