Then perhaps based on user's voting history, it could be calculated how likely a user is to always vote +2/-2, and how likely to actually differentiate between both axes. This could then be used as a multiplier for user's votes on the contribution axis.
It would probably require hiding comment scores though - otherwise it'd be easy to observe how the multiplier changes and game it.