Ok, I think I have understood the quadratic scaling problem now (thanks to @Lauda, @franky1, @jbreher and @-ck), my error was to think that only miners were affected, but as it affects mainly validation, all full nodes are affected and a malicious miner/pool could try to "kill small full nodes" or even smaller mining pools via a spam attack. So my opinion is reinforced that in the case of a block size increase, legacy transactions would have to be restricted by the protocol in some way.