His divisions of four types of target platforms is instructive; the two major axes are free/non-free and Unix-like/other.
“Unix-like” is an awfully big grab-bag, however. The joys of trying to figure out dynamic linking on Mac OS X ought to put to lie the shrill claims of the turtlenecked faithful that it’s just BSD with a shiny GUI (maybe BSD circa 1987, when everybody wanted a magnesium case). I almost dare not even imagine how much fun it is to coax compilers on non-free Unix-like platforms into interpreting modern C++ correctly.
The nice part of free Unix-like platforms is that they’re easy to obtain and install. The monster machine sitting in my other office can run multiple VMs for *BSD and OpenSolaris simultaneously, so testing a patch for portability requires a little bit of system administration and a little bit of discipline to script the process. Testing a non-free platform, Unix-like or not, is much more difficult.
POSIX and free redistribution and source code availability gives us a much better chance of figuring out and fixing those problems than we’d have otherwise. However, that’s no substitute for platform-specific experience – having an OpenBSD VM running doesn’t mean that I automatically know why OpenBSD’s handling of, for example, complex math is different from that of FreeBSD, or how to fix it. Sometimes there’s no substitute for a little elbow-grease from a passionate user of the platform. It’s nice that projects such as KDE actively support it.