Interesting! One argument against this attack is that it’s extremely unlikely that a committee will be a coordinated group among themselves but not have a stake in the rewards of the other validators. Realistically, for any committee attack the be possible, there needs to be an attacker with the ability to control a near-majority of validators, and even though they could increase revenue for the committee, they would lose revenue in the non-committee validators that they control. Still a good argument for keeping the committee size not too low though!
(Edit: just realized that my reply is basically the same as your own last paragraph)