The simulations you refer to operate on an assumption that the distribution of strategic behavior is not correlated to voter alignment. The scenarios I refer to in passing as "weird" are ones that are highly unlikely under such assumptions, and as I note, probably unlikely on the whole.

In order to get truly weird results, one needs to introduce correlation of belief and behavior on the group level, e.g., having some groups of voters that share preferences and are extremely pessimistic, some that are share preferences and are extremely optimistic, and have different prior beliefs about how the election will actually proceed.

At that point, you can construct stories that support going anywhere within the convex hull of non-trivial results. This is, as I said, not the most pressing concern to be had; the more common "worst case" concern is that results could be similar to a plurality vote.

As far as that goes, I would consider your example results from March 2021 neither weird nor dissimilar to a plurality vote.