The law of excluded middle is a law used for the types of propositions that we usually hold to be either true or false, e.g. arithmetic propositions such as "All even numbers are squares." We're well familiar with other types of propositions where it doesn't apply. For instance, it's not either true or false that I am in the living room, since I might be partly inside and partly not; it's not either true or false that there is a god because we'd first have to come to agree what it would mean for there to be a god, which will likely prove difficult; and it's not either true or false that Darth Vader is dead because it depends on future movies.
So there's nothing mysterious about the applicability of the law of excluded middle being restricted. What's confusing in the case that you mention is that at first sight the sentence "This proposition is false" seems like a proposition in the realm of logic, in which we assume the law of excluded middle to hold. But this is an illusion – it's actually a natural-language proposition that uses the fact that natural language has deictics like "this" to establish reference and self-reference. Numbers and logical propositions don't refer to themselves. So the problem isn't in believing in the law of excluded middle for numbers and logic but in misidentifying this natural-language proposition as a purely logical proposition.
Now the fundamental idea that Gödel used to prove his first incompleteness theorem is to make numbers and logical propositions refer to themselves after all, in a sense. By doing this, he showed that a sufficiently powerful formal system is either inconsistent or incomplete, i.e. it's either useless or there are propositions that it can neither prove nor disprove. Thus, there's no law of excluded middle for provability in formal systems – every sufficiently powerful formal system violates this "law". But that doesn't mean that these formally undecidable propositions are neither true nor false – you just need a meta-system that can reason about the first system in order to prove them, because trying to let the system reason about itself would lead to the same sort of self-referential inconsistencies as regarding "This proposition is false." as a logical proposition.
P.S.: As Asaf pointed out, Gödel's first incompleteness theorem applies only to formal systems that are recursively enumberable (i.e. whose axioms and inference rules an algorithm can enumerate).