Ethics are the most probable reason I can think of, and the one I think would make it worth the risk of negative consequences - because there are a lot of negative consequences to unethical behavior, even if they aren't quite as apparent at first glance.
The other options are worlds where a robot's right to self defense is absolutely limited by species (not-human) and circumstance (must-obey). If the robots are people, or are capable of being people, or if somebody somewhere thinks they might someday be capable of being people - then denying the right to self-defense, or self interest has a shiny historical parallel - that of slavery.
People are complex and complicated - to assume that every human life, by the value of being human, must outweigh any robot life, only works if no robot life is ever capable of being a person. And those qualities that make a person, a person are really not so simply defined. Think of different ways sapience can be defined; or the way that the definition of "people" has in the past been limited by skin color, nationality, religion, language, culture, and a lot of stuff we don't think relevant to the definition of people now - and if our definition of people in the future might someday regard organic vs mechanical origin as we regard skin color. Because if we're not really, really sure, or really really right, that robots can never be people, the fact that people can accomplish hard-coding second-class status by virtue of programming (brainwashing) doesn't make it less ethically, morally wrong.
If the robots can't defend themselves even from serious threats and even if the harm or possibility of harm is reasonable to the threat, they are vulnerable. If they're vulnerable, there will be those who take advantage, whatever that might turn out to be. and if we value threat or harm to those guilty, more than those vulnerable because of unchangeable circumstance (what they are, not what they do), with no hope of being otherwise... that says something about us, something that we might not like to hear.
So, letting robots have self-defense is a risk - perhaps as much of a risk as loosening restrictions on any oppressed or enslaved people is. And it probably is a risk, because there will always be people who behave badly (whether or not that means humans only or robots as well). Some people will do so anyway, because they think it's the right thing to do. Some people will think the risk of robots misbehaving is less than the risk of treating potential people poorly. And, if robots can take advantage of, or abuse the loophole... they probably are advanced enough to deserve the right to defend their own lives.
As for placing obedience over any other criteria... the basic assumption that human people, by virtue of being human people with no other restrictions, will always give orders that are moral, ethical, lawful, possible - strikes me as being insane. Putting it on your law-list, and among the top three laws, seems to be making a lot of a-priori assumptions I'm not that comfortable with. Putting it below self-defense and lack of harm to others seems the wisest if it must be added, but yeah. Killbot hellscape, no question.
Moving on - the benefits to society are a little less obvious. Certainly, I think it benefits society to be ethical, or honorable - at least to try. Isn't it said, judge a society by how they treat their most vulnerable? I think a society that will allow slavery is less desirable than one that doesn't. But there's another layer to it, one that's less obvious and more damaging... Most people really don't do well with power. People interacting with beings (people or not) that must be loyal, and must obey, even at risk to themselves - I expect it to go straight to the heads of many people.
Such people would be arrogant, or short tempered because they deal with those who can't say no some of the time, and then deal poorly with those who can. It might encourage stratification and hierarchy, since some roles (robots) are absolutely inflexible, with no disagreement or backtalk or any possibility of changing their circumstance, some people would expect or demand the same from any people "subordinate" to them or somehow in a lesser position. Dealing with people is harder and more flexible than dealing with roles, but it is also easier to treat interactions similarly, either all as roles or all as people. So while some might treat the robots more like people, others will treat people more like robots rather than being flexible, and separating the different kinds of interactions (something kids are particularly bad at).
Having robot-slaves might also interfere with the development of empathy, or mask its lack. Looking at the ways people (especially children) treat animals or insects is often a character reference (including psychological diagnosis, like sociopaths) - less because we value those individual animals, but because it shows how they deal with vulnerability, and lets us suggest how they might deal with those we do value (people) if they end up in a vulnerable position. Robots, who may seem very like people but are outright denied self-defense no matter the provocation - may encourage cruelty, and teach apathy to others. Or it might mask the actions of those genuinely cruel, since they have an outlet that can't defend themselves and must obey. It is a very different lesson than one gets from interacting with animals, who will always defend themselves if provoked enough.
After all that, whether allowing a robot to prioritize its own self defense, will end in a killbot hellscape, a terrifying standoff, or genuine cooperation and peace, all depends on your people. If you assume people are awful, and behave badly, and double bonus if that "people" includes robots, the results are much worse than if people are trying to do the right thing, and make allowances for other, imperfect people trying to do the right thing. If one side is behaving badly, I'd expect the standoff. If people (on both sides) are trying to be ethical, and are willing to work for it, the prognosis is pretty good.