I did make a mistake here regarding the possibility of tactical voting; but it wasn't to just forget Gibbard's theorem. There is a big difference between saying that voting systems are theoretically vulnerable to tactical voting in at least some situations, and saying that they are vulnerable to tactical voting in practice with specific models of voter behavior.

In particular, Condorcet systems are most obviously vulnerable to tactical voting when there is no Condorcet winner, which is always a theoretical possibility, but in these spatial models is extremely unlikely. With more voters, the probability approaches zero for uniform or single-Gaussian models, and about 3% for mixture models. Because lacking a Condorcet winner is so unlikely, even though Condorcet systems are very vulnerable to tactical voting in that specific case, it doesn't really have much of an impact in practice when voters follow this model.