Much more orphans. If you're using POW, the delay in the block relays would be much more significant than the block interval.
If the number of required confirmations remains the same, the security of the merchant accepting the transactions would decreases significantly. As compared to before, less resources is required to attack the chain and double spend your own transaction. I can't think of any significant advantage to lowering the block time to that amount.
Instead of a strict targeted block time, would it be possible to adjust difficulty partly based on orphan rate? So that as orphan rate increases, difficulty would also adjust upward, increasing block time (and lowering orphan rate).
I've always liked this idea.
A fully endogenous solution to block time.
I'm not certain that all of the mining nodes are going to agree on exactly what the orphan rate will be. It is quite possible that some orphan blocks are never relayed to the whole network. This may cause the mining nodes that are not aware of a particular ophan block to mine at a lower difficulty than other mining nodes. And these miners that never had this information would end up getting their block orphaned because their new chain would have less accumulated work than a miner who found a block at the higher difficulty rate.
All the data required to determine the orphan rate would need to be in the longest chain - so all the uncles would need to be referenced. No other data would be taken into account when calculating consensus. So those uncles that did exist but weren't referenced would be ignored, until added to the longest chain.
This way it would always be possible to independently calculate the correct required difficulty from the parent block (and the chain up from there ).