That only happens when a fair much just wasn’t possible. For example if they end up having to match a 6-stack vs 3-stack, a 2-stack and a solo. It’s not like the system is trying to create an unbalanced match but instead using the stats at hand it realizes that the matching isn’t perfect (in my example the 6-stack might have something like a 53% chance of winning).
However - and this is the important bit - what happens then is that the SR gains and losses are adjusted by that predicted win percentage and that is part of the whole foundation of an ELO system like this.
To put it in perspective. Even if the matches were consistently unbalanced so that one team had an 80% predicted win chance the system would still be able to place its players just as correctly as it does today. Under this system and using my example - if two teams with 80% and 20% win chance would face off in 100 games and the stronger team would win 80 of those games and the weaker team would win 20 games everyone would end up exactly at the SR they had before the games. However if the weaker team proved to be better than their MMR indicated and they would win 15 out of 100 games the players in that team would climb as a result - despite losing 85% of all their games.