That being said, I do think the risk of miners publishing "ginormous" blocks if there was no cap at all is small. Based on real data, TradeBlock today
estimated that it would take 137 seconds for a 8 MB block to propagate to half the nodes. By the time a 100 MB spam-block propagated, miners would have found another two and the spam block would be orphaned. The probability of success of such a spam attack is thus inversely proportional to the damage it can cause.
This is an incorrect or at least misleading analysis. The block doesn't need to propagate to 50% of the nodes, only 50% of the hash rate.
In the case of (simplifying for illustration here) two large pools that together constitute 50% of the hash rate, that could mean propagating to
exactly one other node.