More intelligent queue/rating system

Wow arena rating system is based on elo rating system mostly used in chess and initially creaded for it. It's based on the idea that players with equal rating have equial chances to win and a player of higher rating has higher chances to win a lower rated players.
quote from wikipedia: https://en.wikipedia.org/wiki/Elo_rating_system
"The difference in the ratings between two players serves as a predictor of the outcome of a match.
Two players with equal ratings who play against each other are expected to score an equal number of wins.
A player whose rating is 100 points greater than their opponent's is expected to score 64%;
if the difference is 200 points, then the expected score for the stronger player is 76%."
But unlike in chess where the skill of the player (what is what players rating represents) is mostly the only variable, in wow besides players skill a very huge role playes player's comb and his opponents comb.

The proposal below is mostly designed for soloq since it's the most active bracket with huge amount of possible combs but it proably can also be applied to other brackets. (plus soloq by itself is completely custom thing for wow)
So the proposal is to consider both team ratings and the chances of the combs in a game.
For example tsg with pala versus tsg with priest with equial rating assuming tsg with pala has 80% chance to win would give 4.8 rating for win and -19.2 for loss to palatsg team so 4.8*0.8-19.2*0.2=0
Where do you get those chances? You keep the statistics of all games played on hight rated mmr (very simple example: keep statistic on all games where both teams have higher mmr than the rating of player #100 in the ladder) and reward rating change accordingly to the table of statistics.

More detailed information and a possible algorithm-wise realization for this system here:
http://forum.warmane.com/showthread.php?t=346820
I mostly covered the option of forbidding matchups with uneven chances completely but an additional rating change correction is very similar