We already suggested something that could fix this :
do not use simple average of rating, but an adjusted average, taking account of rating dispersion of players.
We suggested two different ways of doing this.
method 1 :
team rating = average rating + k * standard deviation (k a number between 0 and 1)
with k = 0.5 : a team with 1000 and 2000 will get 1750 rating instead of 1500.
method 2 :
team rating = ∆ log(AVERAGE [ 2^(x_1 / ∆), 2^(x_2 / ∆), 2^(x_3 / ∆), 2^(x_4 / ∆) ] ) (where ∆ is a number that can be adapted to our needs)
with ∆ = 400, a 4 player team with 1000,1000,1000,3000 will have a rating above 2200, and a 2 player team with 1000 and 3000 will have 2600 team rating.
these ways of adjust team rating calculation have been suggested in Analyses of the ratings - Spotting the issues - #221 by ReanuKeeves00 BEFORE the last modification (this modification was still a huge improvement, but with flaws)
these ways to calculate team ratings do not affect much teams with players of similar ratings, but will give teams with huge differences of ratings harder opponents than they have now, and thus limiting the current exploit.