If adding a variable changes the actually effect being estimated, then it may indeed have been worth adding.

If it doesn't, they'll just eat away at their precision.