why does the 0 tx block have to come "immediately" after a large block?
They don't. Empty blocks can come after any sized block. But I just showed that F2Pool is
more likely to produce an empty block when the previous block was large, than when the previous block was not large.
This makes sense to me because I expect that for large blocks, there's more time between when F2Pool has
just enough information to begin hashing, and when they have processed the block and sent a new
non-empty blocktemplate to their hashers to work one. If this time is longer, then there's a better chance they get lucky and mine an empty block. See what I mean?
So one could assume those empty block would never be mined or would be orphaned should a competitor be SPV mining.
I'd also assume this phenomena is also only viable while the block subsidy is high enough that transaction fees are inconsequential. In the near future this strategy would most likely be optimized to including the minimum tx's that are low risk high reward to balanced against the risk of loss for the fiew seconds it would take to include in a block.