Indeed, but it is now apparent (to me anyway) that simply increasing the block limit to allow larger and larger blocks to propagate is not necessary, as this is not the optimal long-term solution. The optimal solution takes advantage of the fact that most transactions are already known to most peers before the next block is mined. So, "highly abbreviated" new blocks can be propagated instead. This is beyond mere data compression because it relies on the receiver knowing most of the block contents in advance.

We see in the O(1) thread that there are excellent proposals on the table for block propagation efficiency:
A) short transaction hashes: as in block network coding, and similarly in the optimized block relay (Matt Corallo already has a relay service live)
B) IBLT blocks

Even better, they are compatible such that A can be used within B giving enormous efficiency gains. This must be the long-term goal.

The next question is: Can the max block size be made flexible (for example: a function of the median size of the previous 2016 blocks) as a phase in the process of introducing block propagation efficiency as a consensus change?