There are two ways you could think of bot commands: kinematics or dynamics.
At first, when I was playing around with the new physics, I imagined bot commands to be dynamic: that is, specify a force and have the bot respond appropriately.
However, as I've played with it more, I think a kinematic approach is preferable. That is, specify a desired action and have the amount of effort needed to arrive at that result be calculated.
Mass and radius for bots are highly variable, so having sysvars use forces means that evolution is going to have a harder time to accomplish a comparable task using kinematic controls.
So then, if you view aimdx and aimsx as being angular velocity requests, it becomes pretty easy to implement rotational costs even if you don't have a proper physical system up and running. All you need to do is add another variable to the robot data structure to remember your old aim.
It may seem a little odd to do it like this, where storing, say, 5 into aimdx every cycle means you don't get charged anything, but I think it makes the most sense.