21 TB of data per year is more than manageable. Not by some yet to be developed future tech, but by today's equipment that has a price tag. With 2k$ HIVE that hardware becomes affordable even for backup witnesses.

As for the claim that consensus nodes are able to handle sustained max block traffic, it can be made because state does not really grow with bigger blocks. The things that grow (assuming the same mix of operation types as in mainnet) are number of comments and account related data, but these are subject to optimizations (points 7 and 8), at which point the growth really happens on disk rather than RAM. In reality we should rather expect increase in use of custom jsons, because blogging won't suddenly attract many new users, but third party apps might. Custom jsons leave no mark in state of consensus nodes. Block log growing too big? Split block log and pruning is getting final touches, but is already merged into develop. The only problem with such large block log will be replay, but many people already use a workaround by downloading prepared snapshots rather than performing full replay.

There is a lot of room between current level and max possible traffic, a lot of room between current price and 2k$. It won't suddenly jump between the two leaving us surprised. Also unlike Bitcoin, where changes are hard, Hive has no trouble in adapting. Price increase would be beneficial, because we could afford to finance development from DHF, which is not the case today.