IRV failing to capture voter preferences is not a "corner case". It is absolutely unclear that IRV is better at all than FPTP. See a set of voting simulations with candidates and distributions here for different voting methods:
Alarmingly, IRV is the worst voting system of all just by eyeballing these simulations. The charts it produces are bizarre, and it tends to select for extremists and/or candidates that do not capture the voter's preferences!
It's not just of theoretical interest either, IRV famously spoiled the 2009 Burlington mayoral election, resulting in its repeal: https://en.wikipedia.org/wiki/2009_Burlington_mayoral_electi...
The only good takeaway from IRV picking up steam is that maybe it puts us on the path towards something like approval voting, which is much harder to game, and is much easier for voters to understand (I doubt more than 1 out of 10 people could explain how IRV works).