Why would the Linux kernel developers worry about making their code work on say Microsoft Visual Studio compiler or the IBM xlC compilers?
When you're writing a kernel, you need very precise control over a lot more stuff, like exact memory layout, than you do (generally) in userspace. Such controls are not really accounted for in the C standard (left as implementation defined characteristics for instance), so either some extensions are necessary, or you need to rely on the compiler's quirks.
Sticking with one specific compiler, taking advantage of its extensions, is a rational decision. The code doesn't need to be portable across compilers - it needs to be efficient and portable across different hardware platforms.
