OK, I spent the morning looking into this. Here's what I dogmatically yet provisionally think:
1) Smyth makes it really confusing. I am not sure whether that is because he is just regurgitating Kühner-Gerth without first digesting it, or because he insists on discussing ἑαυτῶν and σφῶν forms at the same time. This is a case where Mastronarde is much easier to use.
2) Mastronarde tells us that the σφῶν forms are indirect pronouns. They don't refer to the subject of their clause.
3) Usually, there are just two clauses, so grammarians will say that the σφῶν forms refer to the subject of the main clause. But when there are three or more clauses, we need to realize that they are going to refer to the subject of the clause one level up. There seems to be lots of needless confusion about this. Better yet, I think it is best just to think of them as referring to the "other guys/things", ie not the first guys/things you might think. The completely unqualified armchair linguist in me thinks that this is all that was going on. The Greeks had the ἑαυτῶν forms when they needed a reflexive pronoun. But when they wanted to refer to the "other guys/things" who are in the picture they pulled out the σφῶν forms.
4) In Th. 5.38, I think the pronoun has to refer to the councils. Fowler gets it right, but he muddies the water when he explains it. Yes, here it is referring to the subject of a dependent verb, but that is only because there are three levels.
5) Graves seems hopelessly confused simply because of the three levels: "[in the other examples] it does refer to the general subject of the sentence, which is not the case here." That's right, because there aren't three levels in the other sentences. He cites Th. 1.20 and 4.113, both of which are unproblematic. 4.113 is interesting because, thanks to our friend osos, the subjects are the same, and the pronoun actually refers outside the whole sentence!
6) What of Smyth's infamous N.2? Well, clearly Thucydides is not using it indirectly. Why is that? Well, I think it is just for emphasis. Normally, one would fear the allies of one's enemy. Here, however, the Athenians are fearing their own allies. So for emphasis Thucydides uses the σφῶν form. We might strenthen this line of argument a bit by noticing that ξυμμάχους is a funny word that has a kind of possessor built in, and maybe just maybe, Thucydides felt that the usual reflexive might only point to that built in possessor. So one would naturally fear the allies of one's enemy; the built in possessor is one's enemy; hence, the reflexive might just be felt to point only at that built in possessor, ie the Spartans; but that is not the case. Now this isn't a matter of syntax or grammar, but just of semantics. And it isn't necessary, but just a case of Thucydides wanting to be clear. Notice also that it is the last word in the phrase, so he has already introduced allies and fear, so he needs to sharply reverse the reader's natural semantic tendency: not the Spartan's (own) allies, but the Athenian's own allies. (I always imagine these Greeks not being able to easily erase, so when they begin their long sentences, like all of us, the don't exactly know how they will end, so strange formulations are in some cases to be explained by semi-desperate attempts to make clear something that perhaps could/should have been made clearer at the outset. Am I wrong about that? Was it easy for them to erase?)
7) I am most interested in particular sentences that are thought to present difficulties for this general line of thinking. I'm sorry, after five hours of thinking about it, I don't see anything very troubling.