I think Erdogan's insight into the game theory behind the block size limit was correct.

But to recognize his insight, I think we need to stop thinking in terms of "valid blocks" and start thinking in terms of "valid transactions."  All blocks that are composed exclusively of valid transactions are valid.  

Now instead of thinking that only Core and XT exist, imagine that there are dozens (and in the future possibly hundreds) of competing implementations of Bitcoin.  Each implementation has its own rules for what block size it will build upon.  From this viewpoint, the "effective limit" is the size of the largest block that's ever been included in the Blockchain.  If a miner wants to create a larger block (e.g., to collect more fees), then he has to weigh the chances that his block is orphaned with his desire to create a larger block.  If we imagine that the block size limit across the network forms some distribution as shown in the chart labelled "NEW THINKING" below, then, since the miner can't be 100% sure what this distribution is, it is rational for him to use the tip-toe method to minimize risk.