Yes we have looked at this and after launch its another area to consider. It requires a different cache mechanism called deterministic cache. So for changing data (say twitter feed) the nodes holding the data can push out in a deterministic way cache copies. So they all copy 1 hop and if that’s not enough it goes one more and so on. The original holders know the hop count of this cache and when an update happens they can push the update Xhops knowing it will update all cached copies.
Then the subscriber/broadcast mechanism can broadcast the message out these hops and the outer edges can send the message to subscribers. The deterministic cache holders can tell they are sending to too many subscribers so they request a cache size increase, the original holders then push this 1 hop further (halving the subscribers to be sent by each node) and so on. There are a couple of nuances (like confirmation a node is really working to hard and not requesting cache for laziness (it balances out anyway)) but fundamentally yes this is all very doable.