One of my test results from 2014 is that the 2 coms team wins exactly as often as the bigger team. Probably this is still true with new income share. Probably the system compensates its own missweighting of 2 com players by distortions in the elo scale (influencing elo of players who tend to get 2 coms).
For the current system and all my new systems I have generalized formulas for giving any team number k any number X_k of extra coms and giving the extra coms player a new weighting 1+w*X_k, where w is a weighting constant for extra coms. In current discussion we only had w=0 or 1, but any 0<=w<=1 is possible. I didn't test those generalizations yet, because currently the system seems to compensate it, which probably causes other errors, though.
Of course restricting the system to only a certain kind of games will improve accuracy for those games as long as there are enough games of this type. But then balancing uneven teams with elo values that are only calculated from even teams with a non-generalized system could worsen uneven team balance. Seperate elos for even and uneven teams could be an improvement, but definitely too circumstantial for a little effect that is again achieved by elo distortions.
Besides proper matchmaking (which requires enough players and should still be possible to circumvent by !forcestart for private games) or AI, a real solution would be either a generalized system or a proof that no big improvement is possible considering the possibility of elo distortions.