"Luck" in bitcoin mining is an Erlang distributed variable. For one block, the canonical "luck" calculation is defined as submitted shares / expected shares. The calculation cares not one whit whether or not the solved block is valid or not. If orphaned blocks were not included, the luck calculation would be artificially low. Orphaned blocks have no affect on how luck is calculated. If orphaned blocks were not included, expected shares / accepted shares would no longer be Erlang distributed, and "luck" calculations averaging more than one block could not be made.
So instead, I'm using the "bitcoin per gigashare" metric as a measure of income that includes the effect of both luck and orphaned blocks. Reward method effects and fees are not included.
Summary: If you want to look at the canonical luck (the Erlang distributed average of submitted shares / expected shares) look at the "luck" column. If you want to see the effect of both "luck" and orphaned blocks, look at the "bitcoin per gigashare" column.
BTW, the comments in the private email above were about the old table. The quoted table is one of the new tables which I developed after taking on board your input.
You'd expect more short rounds and clusters of them just because of the shape of the CDF, and observations just glancing at blockchain.info every once in awhile are obviously subject to huge reporting bias, but a lot of people seem to complain about GHash doing a withholding attack.
I think it'd be interesting to get a comparison of block finding time for the different major pools, though I have no idea what type of distribution you'd expect around their average generation time.
