Busybox's include/platform.h was added on my watch to support the guy building against libgloss and the people who wanted to build against MacOS X and so on.
That said, board support packages usually provide kernel source because it's one package and you usually have to rebuild it to tailor it to the hardware. They don't necessarily rebuild the base userspace.
For example, at my day job the "bringup" department I'm working for has built kernels for three different new product boards since I got here. I build kernels every day. And each time they use a binary arago root filesystem tarball that hasn't been recompiled since I got here, with a toolchain they got from code sourcery in 2009. (The bringup department then hands said kernels off to the Android guys, who replace the root filesystem with Android stuff they get from Google, and layer their own stuff on top of it as android packages. Neither the kernels nor root filesystems actually _ship_, I'm just trying to make the hardware work and then they put android on it.)
Google's already excluding all GPL code from userspace, and android developers are happy to go along with it. I'm building a package the android guys might actually get to _use_. Busybox has existed since before android shipped, and doesn't get used on android, and isn't going to. Arguing whether or not it _should_, or how they'll come around if we wait for the sun to go out, has nothing to do with reality.
Linus Torvalds stopped waiting for the android developers to come around to his way of thinking, and started merging their code. I've usually considered Linus a good model. (Remember how he used a non-GPL license for Sparse because he was fed up with the FSF? I suspect git was only GPLv2 to quiet the "oh no bitkeeper"! hysteria, but haven't asked.)
I keep hearing people go "what, are they going to rewrite the kernel next if we make a big enough fuss about it"? And I keep going "MacOS X is BSD based became the most profitable company in the world using that; you think Google can't switch to that in a single development cycle if they really wanted to?"
Linux has stayed unified because of _LINUS_. Thinking that "oh it can't fork, it's GPL"... Android kernel anyone?
Red Hat almost standardized on Alan Cox's kernel back when Linus overloaded and spent months dropping patches (anybody remember my old "Patch Penguin" rant?), and Alan took a year off to go back to grad school to FORCE everybody to work with Linus (who took up bitkeeper to solve the scalability issue). The license has _helped_ keep Linux unified, but it's the community that's actually done it. The community of kernel developers who _REJECTED_ GPLv3, sidelined Richard Stallman, and never BOTHERED to launch widespread license enforcement lawsuits because they are a BAD IDEA. There's some saber rattling, but suing potential allies generally doesn't bring them closer to you. This isn't a defeat = friendship world.
Me, I reinvent the wheel a lot (hence working on busybox). I had to prove for myself that they're a bad idea. So I ran the experiment, and I reported the results, and whadaya know: It's a bad idea.
(I can't help it if other people want to repeat my mistakes and insist on finding out for themselves that fire is hot, but telling the guy with firsthand experience that you know more about it than he does gets old after a while.)