The first thing you need to understand that it's not just a matter of the majority of miners for a hard fork.... it's got to be pretty much everybody.
Quite true. In fact even more so because "old" protocol nodes will only accept small blocks, while the "new" protocol nodes will accept either small (<1MB) or large (>1MB) blocks. Thus all blocks produced by old miners will be accepted by the new ones as valid, even when there's an extra 500KB of transactions waiting in line to be published.
You'd need like a >90%, simultaneous switch to avoid total chaos. In that case substantially all the blocks published would be >1MB, and the old protocol miners wouldn't be able to keep up. If normal nodes switched at the same time, they would start pushing transactions that old-protocol clients / miners would lose track of. It seems very likely that when / if the change takes place, blocks will have been at the 1MB limit for some time and the end of the limit would immediately result in 1.5MB blocks, so it would have to be coordinated well in advance.
It's not that bad. If the larger block miners are >50%, they will build off the longest valid chain, so they will ignore the smaller block miners blocks since they have a lower difficulty. If the smaller block miners are >50%, they will always have the longest chain and no large blocks will ever survive reorganization for more than a couple confirmations. Block headers contain the hash of the previous block, so once your chain forks, the blocks built after the first split block are not compatible with the other chain.