Hopefully ghash.io is not testing out the selfish mining strategy...
I'm more inclined to say incompetence over malice. I honestly think they don't have the slightest clue what they're doing when it comes to setting up a pool and bitcoind node, as demonstrated by their almost daily outages of either their public pool or private mining side (or both).
On my opinion the problem is that most of their hashing power is in locations with low cost electricity and space, which is usually in locations with bad connectivity (at the end of the world and a bit further). Because of that each datacenter is working on it's own mostly, with just few outside connections to synchronize with the network.
So you mean that GHash.IO are could be causing more orphan races than usual, rather than winning more orphan races than usual? Or both?
Am I right in assuming that this is only an issue because they have such a large chunk of the pie?
I think it's both (causing and winning), because of large chunk
separated from badly connected to the network.
It was similar when ASICminer had a big chunk in a single location with good connectivity - GHash.IO have it in few with bad connectivity, so the problem appeared a bit earlier.
There should be no single entity with much power (no more than 25%), but there is also another reason causing the big number of orphans - the rate is 9.63 blocks per hour, which is also 30+% over the normal rate for the network. Once it drops below 20% the things will normalize.