The Supreme Court too was confused on this, calling the entire code base "the API" in it's decision.

The [actual] API is 'the most valuable' part of the code, true, but it's also the part least protected by copyright. "Factual" descriptions/specifications are not in fact copyrighted, and the right to make a reverse-engineered clone of something was already established under US copyright. If you want to for instance make a specification and say that nobody can implement it without your permission -- you can't actually do that under copyright law (you might be able to under patent law, in some cases).

If I were the judge, I would have decided the case based on that -- in favor of Google. That it's "fair use" to copy the API in fact regardless of what % of the work it was, precisely because cloning a specification is fair use, copyright is meant to protect a specific fixed expression (in the case of software that means the implementation) it's not supposed to prevent you from creating a compatible clone in the first place, and never has been interpreted that way before this case, it was a mistake, and you don't even need to decide what % of the code "the API" (the headers/method signatures) are to get there.

But that's not actually what happened in the US courts, the judges involved seemed (to my reading) to be very confused about these things, and not to really understand what an "API" (as specification) was, or it's role in software. They decided on different grounds. To me making the law pretty incoherent and unclear at this point. So it goes.

Also, "how difficult to create" something is has very little to do with it's protection under copyright in US law, generally, whether you think it should or not, not how the law works.