This is by definition; honest voting is not rational (in the narrow sense of the term).

Not necessarily; I think it's usually more of a prisoner's dilemma situation: if you optimize for getting what you want, then the overall outcome is worse.

For example, in the results Borda count is much worse when everyone strategic votes.

The new results contrast with an older simulation (graph shown down the page): https://electionscience.org/library/tactical-voting-basics/ Those results show honest voting always being (globally) better. That's one reason why I find the new results surprising.

Thanks for broadening the discussion beyond low-trust, purely self-interested rationality.