Are you suggesting moving that down the ladder and not having that data everywhere? I.e. if someone want to see an old post, there would be one extra step required to load the data (so cdn, cassandra, now subset of postgres with not so old data, "full" postgres). I think facebook does something similar, but they really have to, considering their size.
In terms of status updates (ie, stories which may mention check-ins or photos or similar, but not Facebook Messages), before Facebook's Timeline launch, there were multiple stores of data depending on age. With Timeline, all the different versions of data over all ages were put back together into a single (logical) store. More about that process at:
