What about this, don't count any fees, just declare invalid any block in which OP_RETURN and BIP 9 (user's and miner's) flags are different?
I don't see obvious holes/abuses with this approach.However, it would only incentivize miners to change their BIP 9 flags when proportion of blocks signalling certain flag is significantly less than proportion of transactions demanding that flag. For example, miners would be incentivized to signal BIP_N if 70% of txes demand BIP_N, but only 30% of blocks signal it. Now if 90% of hashpower started to signal BIP_N, but 30% of txes demand BIP_M, miners will be incentivized to switch signalling to BIP_M, until more or less ratio BIP_N_blocks/BIP_M_blocks = BIP_N_txes/BIP_M_txes.
Also, this approach wouldn't work without majority of users actively expressing their preference through their transactions, despite sometimes having to pay higher fees and experiencing additional delays in confirmation times. I guess majority of users would continue generating transactions of default type, because they care more about fees and confirmation time than about something they have only vague idea about. So most of time we probably will be having like 90% default txes, 7% BIP_N txes, 3% BIP_M txes, what doesn't create any incentives to change signalling at all.
Nonetheless, in my opinion the idea is interesting.
