Just about every language NOW has a good FFI to C. Subtract several years for when that essay was written, then subtract ten years for when the statement is set in, and you'll encounter a different world.
Today there's no great reason not to write your desktop app in PyQT or Clojure or on top of XULRunner. (Although I'd point out that surprisingly few people have noticed this!) ViaWeb was born into a different world. I didn't do much desktop programming in the 1996 timeframe, but you did not have many choices, especially if you were a student that couldn't drop ~$1000 2009-dollars into a programming language environment. (Now my new supercomputer-laptop costs less than that and superior environments are free. Vive la progress!) A couple that weren't C and, as it turns out, all doomed to be unsupported within a few years anyhow, like Delphi. I talk about 1996 because it's when I got into the game; the essay is a talk given in 2001 which means "ten years ago" is 1991.
Another reason is documentation. You'll have an easier time finding documentation/examples for the main language of the OS.
Another reason: having dependencies that aren't available on the target OS does impose additional development costs (if you don't want to give the user the burden of installing those dependencies). And if the distribution is over the Internet, it also means a bigger download size and extra harddisk space. These things may or may not matter, but imagine a simple feedreader that requires a 30MB download and 200MB of free disk space.
The main language of the OS doesn't necessarily have to be C/C++. It's the language that's best supported and encouraged by the company/community that maintains the OS (although it's currently not possible to write system drivers in C# on Windows, work is being done to lift some device drivers in user mode, which would make a good fit for .NET)