Well, here goes:
1. "Ethical" means a lot of things to a lot of people. Mostly it's a licence to inflict your opinions of other people on them at length by considering them axiomatically correct. It's a human sort of problem, the kind people without any solutions to real problems love to whinge about. So, in that spirit, I think it may be best to point out why paywalls are necessary and leave the ethics to people for whom that sort of problem is important.
Paywalls exist first and foremost out of economic necessity. As long as servers and printing and bandwidth all cost money and curating the peer review process demands specialized expertise, someone has to pay journals to exist so they can pay their employees and running costs. Now, you can say government should do that, because of course "they" should and science should be free to everyone, but do bear in mind that what government funds, politicians try to control. That's how you get things like Lysenkoism: you let politicians think that they can buy their own facts in lieu of real ones, instead of in addition to them. The current drive to maximize impact factor is at least biased towards useful papers rather than convenient ones. So, if not government/"them", we can either ask scientists to pay to get their research out or we can ask industry and academia to pay to read it. The latter two groups have much bigger and more suitably segmented budgets than the former; publication fees still exist, but at least they aren't several thousands of dollars for low-tier journals like open-access can be. In that sense, then, paywalls are, to my mind, the least bad option to keep journals running, because they foist the economically necessary costs on people who are at least biased towards genuinely useful results and able to pay without sacrificing research capacity.
But there's another reason paywalls are useful: they make it harder to promulgate plausible misinformation. People have an inherent drive to make sense of their world, to see patterns and try to place individual observations into a larger context. They're much less driven to validate the results of their curiosity, in part because it's a long, boring process that might not seem as necessary as the million other demands on their time, so people generally stop being curious when they have an answer that makes sense to them. That gap spews forth an endless assortment of homeopaths and crystal healers and fortune tellers and mediums and Biblical literalists and other superficially plausible alternatives to reality, many of which are more readily believed because they're simple and appealing and nice. Open the archives and let the public read whatever they like, and you also invite these people to quote-mine and misinterpret data to suit their ends; further, you economically incentivize them to do so, in order to dazzle with spurious references anyone they can't convince with appealing nonsense. I'm not saying the public is too dumb or too credulous to respond with their own research; I'm saying they literally do not have the time to sort through the resulting deluge of lunacy because they have jobs and demands on their time. There's a level of plausible nonsense you'll believe, same as anyone, and there's a whole lot of people with a financial interest in finding it and selling it to you who will read more literature than you have time for in pursuit of something easier to claim than it is to disprove.
And that brings me to my third and snobbiest point: pick a random member of the public, drop a paper in their lap, and the understanding they reach of its contents will probably be materially wrong. They're written in what is, for all intents and purposes, a dialect, one as different from its root language as American is from British. This is why you get people dismissing evolution as "only a theory" or trying to claim that a p-value < 0.05 is infinitely probative of whatever they want to believe. When you add in the subtle statistics we do and the way we code things, there's a lot of room for even honest error -- and you can see those errors being made everywhere, even when the training to avoid making them is freely available, because it's just not worth it even for self-proclaimed intellectuals to acquire the bare minimum level of understanding required to parse data correctly. I'm not saying that's a bad choice to make. It's not; learning things you can't use is a luxury, and I'd no more expect everyone to understand science than I would expect them all to learn how to responsibly handle radioactive materials or write in cuneiform. It's just not worth it for a lot of people, but it does mean that they're not going to reliably be any better-informed for having scientific papers available, not least because there's a huge amount of text to sort through. We don't spend years in college and graduate school and postdoctoral fellowships and so forth for nothing; we're acquiring a specialized skillset for a specialized task, and assimilating new and complex data in a way that lets us draw useful, falsifiable, relevant conclusions is part of that skillset. It's an easy part to think you have, too, which is part of why people don't believe how much checking we do.
So in other words, yes, I think it's generally in line with what we're supposed to be doing to restrict access to information the public hasn't been trained to use properly, and I also think it's ethical to pay journals out of the deepest (functional) pockets we can find in the process. Paywalls do both; they help keep the information out of the hands of people who are more driven to misuse it than their audiences are to understand it and they keep peer review properly incentivized, albeit in a crude way. None of this means the public is stupid or easily duped. It just means that people do what they're paid to do, and right now that doesn't lead people to do things with free scientific data that are, on balance, in line with our obligation to educate the public.
2. I don't like any of the ones I've seen in isolation. There are fixes, but they're comprehensive ones.
3. N/A. I don't want scientific knowledge proliferated without proper safeguards in place against its misunderstanding and misuse, and the public has little incentive to take advantage of those as it currently stand.
4. Oh, definitely worse. Pirates zero in on whatever papers appeal to them, rip them away from their proper context and plaster them where people have no hope of finding that context. Regardless of how we release data, it must be controlled, and with the background at the very least available. Doing it haphazardly helps no one.
One final point that should probably be made: Your questions suggest you're thinking about this problem in human terms, kind of like how weird has leapt from villain to villain in this thread to explain why he's been unjustly denied something to which he feels entitled. That skips a step. Before we ask what is "ethical", we should ask what is necessary and possible and maximally efficacious, and what must be changed to change those assumptions and the resultant possibility space. People spill mountains of ink telling themselves elaborate stories about how much richer and more popular and better liked they and people like them should be if only the world were just, all the free things they deserve, and what everyone else should be doing to make it so. Ask first what is and what can be, and why. Too often people come at problems like this ethically, become convinced their way is perfect, and spend their time grousing about how evil everyone else is. That's not helpful.
I don't like paywalls as a concept, but I do acknowledge that the conditions that make them beneficial are tied into too many things for it to be worthwhile to cry out against paywalls alone. We can't just switch them off at will. We didn't switch them on at will, but rather at need. It may help, as you form your opinions of this sort of thing, to keep those dependencies in mind.