That's a great point, thank you. I should have tried to describe a more subtle effect where someone sees the policies of the far left candidate and it taints their view of the Democrats, who they now view with suspicion, fearing the Democrats might secretly support the same extreme policies. Alternatively, a Republican voter might see the Overton window shift and consider the Democrats a more centrist party, and so drop their support for the Republicans.
The problem is, these parties (and candidates) aren't spherical cows in a vacuum, and neither are voters. Voters have imperfect information and have to try to make predictions about which politicians are lying, and what direction the mood of the country is moving; so the addition of another party into an electoral contest can quite reasonably cause someone to change how they view the existing parties and even flip their support.
You may be interested to hear about the Gibbard–Satterthwaite theorem then. Any non-dictatorial system with more than two alternatives is subject to tactical voting. Whoops!
It does apply more broadly than Arrow's, but my understanding is that even though a tactic will exist for every voting system, that tactic may require perfect information or at least isn't guaranteed to be actually exploitable by any voters taking part.
If a tactic is so complicated and likely to backfire that no one ever attempts it (or the people that try it, on average, end up harming their cause) then I think that the threat of tactical voting is insignificant.
What I don't know, unfortunately, is which voting system is best in terms of ease of voting honestly and difficulty of voting tactically. Determining that may involve knowing things like how people respond to polling data, or to one party telling their followers to fill out their ballots in a certain way, which might be hard to model.