I should be more clear. Merely because a header file describes a kernel interface doesn't mean that it's not copyrightable on an interoperability theory. It's copyrightable merely if the description itself is sufficiently creative. One way it would not be creative is if the expression is dictated by efficiency concerns, or general engineering best practice. But it's not lacking in creativity merely because it's describing some other code, and therefore hemmed in by the necessity of describing that other code. Even if the header is perfectly dictated by what it's describing (i.e. there's only one way to create the header), if what it's describing is copyrightable than so is the header.
Otherwise, a table of contents wouldn't be copyrightable merely because it's describing the book. But if each chapter heading has a creative aphorism, so does the table of contents; therefore the table of contents would be copyrightable.
Now, if you processed the table of contents, stripped out the chapter headings and were left only with a table of chapter numbers and pages, you would be solid. But that is not what Google is doing. If you think that by stripping out the comments all the copyrightable expression is gone, then you're implying that huge bodies of source code are not copyrightable by themselves, no matter what. That's a dangerous precedent.
Note that this isn't like the SCO case at all. That's a false comparison on so many levels.