It's about 1 in 600.5001389 chance a block will be found within a second following another. However a block find faster than the network latency does not always result in an orphan - it is not an orphan if the same miner also found the following block.

That is the problem with orphans, they reduce the strength of decentralized mining against attack, by favoring larger miners and discarding proof-of-works that would otherwise strengthen the difficulty. An extreme demonstration of this was just had on testnet after a reset to difficulty 1 vs difficulty 100k hashrate: Even holding 1% of the network hashrate it was impossible to get your block find published, because the largest miner was finding blocks at nearly one per second and building upon their own chain even though they didn't have a majority hashrate nor were they running an "attack" client.

An attacker can enhance the chance of a malicious block acceptance by not including irrelevant transactions. This is in addition to a 51% attack actually becoming a 49.83% attack with a one second delay between legitimate miners. There is already a multisecond delay in pooled mining between the pool software learning of a new block from bitcoin, and pushing it out to miners, and their miner software flushing the work and getting the new block hashing on hardware.