I was led to believe the purpose of slow (long) confirmations was to reduce the risk of an attacker mining an alternative chain in secret (to be revealed later as a hijack and a double spend), by making it unlikely to be able to sustain a high hashrate (multiplied by "luck"?)
For a 12 second block time in Bitcoin, the network would likely collapse down to just the biggest pool, as everybody else would have a disproportionately high number of orphan blocks. Sensible miners would need to move to the biggest pool in order to avoid a significant double-digit percentage losses.
> I thought the risk of a split chain was tied mostly to the wall clock, not the block count?
This is a correct statement to make. The number of blocks considered to be safe is completely irrelevant, and only has the effect of denoting the time on the wall. An hour for 6 confirmations in Bitcoin wasn't chosen so much because it was 6 blocks long, but because it is a reasonable tradeoff between convenience and security. If the block time had been 5 minutes you would need 12 to get the same level of confidence in a transaction. I don't understand why the Ethereum developers appear to think otherwise.