I understand Stef's argument that the existence of a bizarre exception doesn't necessarily invalidate a general moral principle in the same way the existence of a two-headed horse doesn't require biology to redefine "horse."
What I don't think I've heard him address is the more fundamental point I think some critics who raise such hypotheticals are getting at:
There's an old joke: "A man asks a woman: 'hey, would you sleep with me for a million dollars?' She says, 'sure.' Then he says, 'okay, would you sleep with me for a hundred dollars?' She slaps him and says 'what kind of a person do you think I am?' The man says, 'We've already established the principle, now we're just haggling over the particulars.'"
That is, if you posit some moral absolute like, "it's always wrong to initiate coercion," then, so long as it is truly absolute, you can stand on the principle, just as the woman in the example might have stood on the principle "it is wrong to sleep with people for money." However, once you accept the idea that the principle is malleable in extreme cases, like, "don't initiate coercion unless it's really, really important," then it stops being a "principle" and instead becomes more of a guideline, the borders of which are open to negotiation. In the joke, for example, the woman already admitted that she wasn't really working with an absolute principle "it's wrong to sleep with someone for money"; instead, she was working with a general guideline: "it's wrong to sleep with someone for money, unless it's a whole lot of money." But once she's conceeded that, then it's totally subjective as to where the cuttoff should be: "is it wrong to sleep with someone for 100,000? 200,000?" Maybe if you're really hard up, it's okay to sleep with someone for 100 dollars.
By extension, if we concede there are exceptions to the non-agression principle, such as "it's okay to coerce if it's really, really important," or "it's okay to coerce if it saves somebody's life" (I would not include, "it's okay to use violence in self-defense" among "exceptions," since the principle is generally understood as prohibiting initiation of force, not all force), then, we've established that the "principle" is actually somewhat flexible and open to negotiation.
But if we conced that the non-agression principle is open to negotiation, then how do we draw an objective line in the sand where we say "this is not important enough to justify coercion," or "that's too much coercion, regardless of the necessity." If we concede coercion's okay with stealing a penny to save the world, then how do we draw a hard and fast line between that and stealing a thousand dollars from a millionaire to give it to a poor person? Or between stealing a thousand from a millionaire to give to a poor person and a generalized system of taxation-funded welfare? I may argue that in A, we need to make an exception, but in situation B, that's too much, or that a penny to save the world is a good time to make an exception, but a whole tax-funded welfare system is going too far, but once we've conceded the principle, we can no longer cite it as an objective standard, but instead are thrown into the messy particulars of arguing why "this much coercion is okay, but that much coercion goes too far."
Personally, my only take on it is to say, "it's still wrong from the perspective of a strict interpretation of ethics to steal a penny to save the world, but in real life, everyone works with a combination of ethical and practial/utilitarian considerations--no one acts in strict accordance to ethical principles or strict utility maximization." I think Stef is implicitly making a similar concession when he says ethical rules are more like biological definitions (the boundaries of which are negotiable and porous), than like axioms of physics. But once we concede that the boundaries of ethical rules are negotiable, doesn't that leave the door open to negotiation as to what degree we should stand on principle and to what degree we should be utilitarian? Once we concede we have to make that principle-utility negotiation, how can we claim objectivity?