It would be a risky strategy because those commenters could get banned.

Risky for real commenters who care about their presence on HN. Effectively consequence-free for astroturfers.

New accounts can be handled differently by the flame war flamewar detector than established accounts.

That is, if this is a actually problem and not only a theoretical one.

Not a big problem if you grow a constant number of expendable accounts.

If I were an astroturfer, grooming accounts would be part of my daily routine. I'd probably automate it too.

But unlike flagging, there is no minimum karma requirement to get involved in a flame war, right?

I think that maintaining a community like this is really a dynamic process, so while it's important to make note of bad incentive situations for the future, it's not necessarily the best choice to correct them prematurely. In theory, your reasoning sounds right. In practice, that's probably not happening right now, so the mechanism still serves its purpose with only the (presumably considered acceptable) expected collateral damage.

The important thing with a situation like that is to be ready to kill the feature as soon as people learn how to exploit it. It's one thing to see that it's theoretically exploitable. It's another to actually figure out how to selectively engineer flame wars to make a significant dent in the front page.