Just to finish up a topic from yesterday (and please read that post first to understand this one): For any type of halachic proof we need to analyze whether that proof tells us something about the metziyus, the facts of the case, or whether the proof tells us something about the din, the halachic conclusion that we should draw. R’ Elchanan (Kovetz He’oros, Yevamos #71) distinguishes between proof that is derived from eidus and proof that comes from rov.
Eidus establishes fact. If two witnesses tell us that Reuvain is dead and we conclude that his wife is permitted to remarry, we have a big problem if Reuvain shows up a few days later alive and well. The halachic conclusion drawn on the basis of eidus is no better than the underlying metziyus it rests on; if our facts are wrong, our conclusion is erroneous.
Contrast that with rov. If a nice trief steak gets mixed up in a display case with two kosher steaks, according to some Rishonim you can actually eat all three steaks. How can that be – the metziyus, the reality, is that one steak is definitely treif!? The answer is that rov does not tell us anything about reality; it tells us about the halachic system. Within the halachic system the definition of kosher is based on rov; if a majority of the meat is kosher, all the meat is considered acceptable.
I would interpret Tosfos in Nazir 58a using this same idea. The rule of safeik tumah b’reshus hayachid helps determine metziyus. Where that determination would lead to two mutually exclusive possibilities being simultaneously true, the rule cannot apply. If a witness testifies that one nazir among two became tamei but he cannot identify which one it was, declaring both nezirim tamei or tahor would contradict realtity. However, the rule of chazakah operates differently; it determines halachic rules, not metziyus. Even where contradicted by reality, chazakah still allows us to declare both nezirim tamei or tahor.