Your suggestion effectively allows the miners to decide the blocksize. It would be trivial and cost-free for the miners to either send many transactions to themselves with an arbitrary transaction fee, or to include fewer transactions than is economically logical.
Actually, there is the cost of opportunity because miners don't get paid mining their own transactions. Also, I don't see why a miner would purposely raise his own storage cost...
A miner does not get paid to confirm his own transactions, but he also does not pay anything to confirm his own transaction. If the maximum block size is based on the last
x number of blocks, including additional transactions will increase the maximum block size in the future. There is some opportunity cost, in the form of the greater chance that a block will be orphaned when it includes an additional transaction.
The additional storage cost of including a transaction is close to zero.
These discussions could potentially lead to good ideas.
One idea that i could think is determine block size limit based on hardware/internet growth. There's one BIP about it which is, BIP 103 Block size following technological growth (
https://github.com/bitcoin/bips/blob/master/bip-0103.mediawiki).
The hard part are,
1. Determine the proper percentage for block size limit growth. The BIP mention 17.7% growth/year.
2. Until when the the limit growth happen? The BIP mention until July 2063, but i doubt today's growth will reflect growth in next 42 years.
That BIP might be reasonable. There does need to be a vibrant fee market, or else the miners will bid down transaction fees to zero, which creates other problems.
If something like this BIP were to be implemented, I would suggest the automatic growth be limited to 10 years, and after 5 years, a new hard fork can be implemented (if there is consensus) to extend the growth indefinitely.