Recent works (e.g., [1,2]) have identified incentive misalignments due to EIP-1559 base fee mechanism. In a nutshell, these benefit sophisticated users who are incentivized to bribe validators, and reap benefits on the expense of slow-reacting/naive users. Although the overall risk to the Ethereum network from these attack does not seem to be (currently) severe, there is a fix to the issue: adding past-weighted averaging to the base-fee calculation formula. By using a geometric series for the weights, a favorable tradeoff is achieved in which the adaptation delay is (negligibly) small and while the attack incentive is significantly reduced.
It seems so, in spite of EIP-1559 itself was designed to improve fee predictability and reduce the problem of fee auctioning. And your brilliant idea of improvements to the base fee mechanism could help mitigate any existing incentive misalignments or attack vectors.
I don’t know of a forensic research on the topic. It would be particularly interesting, however, if it would discover correlation between different proposers that might be working together.