Another intuitive, no-math (and I believe best) strategy could be as follows:

>! The prisoner that gets steak the first night should always vote 2 (Don't know).
>! The other two prisoners that get fish tacos the first night should vote
>! 2 until they get steak for the first time, then vote 1 (Steaks) every night from
>! then on.

This ensures that

 * The majority won't vote 1 (Steaks) when they would be wrong.
 * The majority will vote 1 the first night they all had steak.