The idea behind this is that if we get a stale block a few seconds after the network has seen a block, we'll switch to it instead.
Our chances of winning the orphan race then end up being directly related to our % size of the bitcoin network (and getting a double block), since no other pool will be working on our stale block.
Do the bitmain-controlled pools do this too? I imagine this could be more useful for them since combined they control such a large % of hash power.