In fact the more I think about this, the more I see it as a good idea.
Tack on to e.g. the DNS seeder/peer discover mechanism a fee suggester/oracle. By social contract it will just be @aiyadt’s fee proposal (actually maybe make it change slightly periodically to test that people are using it). But if there is a DOS problem we can the return it to the original fee.
And then later we can make it pick fees via e.g. EIP-1559 as you suggest. And then after that we can move it into actual consensus at some point
it gets us progress towards an end we need AND mitigates the one downslide of lower fees: cheeper DOS that really is cheap enough it might attract trolls. Remember, the other DoS vectors require some skill and effort to get code to do them. Just scripting sending TXs to yourself really does not.