One suggestion. Is it worth mentioning IBLT because by the time 16MB blocks are being mined they will likely be taking up just 500KB of bandwidth overhead to propagate. Obviously the real-time unconfirmed transaction overhead does remain. and is large long-term.
IBLT takes a large spike in needed bandwidth that occurs at the time a block is found and spreads it out over the entire time the block is being mined.
If you're already talking about bandwidth requirements in terms of 10 minutes averages, then introducing IBLT doesn't appreciably change anything.
That's false. IBLT almost halves the bandwidth requirement. Currently transactions are sent twice: once when broadcast and once as part of a block when mined. IBLT reduces the bandwidth requirement of the latter to a constant.