I assume you mean more broadly on op return and not the two implementations.

Mine? lol, they don't matter and I haven't done enough considering to have a strong opinion yet. I find myself agreeing with any well laid out argument made at this point (very uncool, I know).

ATM lean toward being annoyed with core. Leave the default alone, and keep option available. Faster block propagation times seem like a more sound argument than fee estimation for what they're doing (honestly don't personally care about fee estimation right now). But haven't seen any hard numbers on the impact of fragmented mempool on block prop speed. Is it minor, very big? How many seconds we taking about on average?