That imposes a huge amount of risk on the network - if we measure the constants wrong, then it would be easy to accidentally hit an equilibrium where the interest rate is very low and no one participates. Also, in the long term, when transaction fees dominate there’s a de-facto return component that’s a simple inverse of total deposits, so we can’t count on the reward schedule too much.
I think it’s better to just have explicit censorship detection, a la Censorship rejection through "suspicion scores"