As I responded to Wind_FURY, I don't see what would be the incentive for a miner to play with the block size limit (assuming they have enough hashrate to do that);
If they increase the limit then the transaction fees will go down.
If they decrease the limit then the transaction fees will go up but they can validate fewer transactions.
If the block size limit is lowered, the cost per transaction increases. There is the potential that transaction fees in total will increase more than the block size limit will decrease. So lowering the limit would increase total transaction fee revenue.
As Danny explained, a miner can send a large (in terms of value) transaction to himself with zero transaction fee, effectively lowering the average transaction fee for that block at a very low cost.
If there were a situation in which bitcoin
must have a dynamic block size limit, it would be superior to have the block size limit based on a sat/vByte basis. If the cost to include 200 bytes in a block becomes too high, the block size limit will increase, and if the cost to include 200 bytes in a block is too low, the maximum block size will decrease. This would make the mechanism for changing the block size to be consistent with how transactions are priced today. It would also make the manipulation as described by Danny less effective, although said manipulation would still be possible, as a miner could fill 30% of his block with zero-fee transactions to himself (or with transactions whose on-chain fee is zero, one party to the transaction has paid the miner via some off-chain means).