> Maybe, but I get the impression that overall the addition of a true runoff tends to make things better. For example, this guy compared a number of voting methods using his preferred metric, Bayesian regret; note that when a method and a variant with a runoff added to it are compared, the runoff tends to do slightly better

I suspect that result is caused by voters being misinformed about how IRV works. It would be interesting to see IRV compared to Exhaustive Ballot, which is like the Two-Round system, but with up to n-1 rounds for n candidates. The last-placed candidate is eliminated each round until there are a majority of votes for one candidate, just like IRV.