the reason for each individual miner to create supersized blocks is not that he hopes others will do the same. It's simply because it's more profitable for him to do so, regardless of what the others do. This is not the prisoner's dilemma, where there is a conflict between selfish desires and the greater good.
I apologize for making this even more academic, however this is a complex proposal and there will always be games and ways of analyzing this which are analogous to the prisoner's dilemma. For example a large miner (A) could consider that producing a smaller block leaves more transactions in the meme pool, such that another miner (B) would be incentivised to produce a larger block and pay a larger penalty and this would increase the rollover pool, which large miner A could collect.
Miners want to maximize their (tx fees - X * penalty) + collection +- X * impact the decision has on other miners block sizes and the associated penalty's
X = projected share of the network hashing power
That's not the point. What prevents the competition from pulling from the pool without contributing to it? I understand that it's in the interest of a massive node or cartel of nodes to mine large blocks, but only if they aren't undermined by their competitors, which can pull from the pool without paying any penalty themselves.
Nothing prevents miners leaving pools. Is that not desirable? This could potentially be a great characteristic of Meni's proposal. Undermining a large cartel of miners in favor of smaller miners seems great. (I don't know if Meni thought of this when he made the proposal, or it was kind of luck?)
tl;dr version:
With floating block limits + rollover penalties:
mining cartel tries to artificially inflate blocks => they must subsidize smaller miners with penalties
Bitcoin experiences genuine, long-term growth => miners unanimously include more transactions => block sizes will increase
I think this potentially sounds like a good idea.