You can run it on a remote computer (or a VM) and debug it from an OS X front-end that connects to the remote-running debugger over the network. It's a pretty common technique, gdb has had it for, what, fiftenn? Twenty years?
Maintaining two separate development environments, on two operating systems that you have to boot into separately is very annoying. Developing for an operating system other than the one you're running is about as old as the personal computer, I think :).
The remote computer is often good enough to run an application but way too slow to productively compile a large codebase on it. Besides, you don't need to maintain two development setups at once.
This is a niche situation only because Microsoft's tools make it a niche situation. It's pretty much how all embedded software on this planet is developed (except that the network is replaced by a serial connection on the lower end of the performance spectrum, of course). In nix world, it's a super-common situation; e.g. at $work, where I write software for networking equipment, no one seriously expects that I'm going to compile stuff on the routers that ends up running that software. Most of them are capable of the required heavy lifting (they have CPUs in the same performance class as a pretty beefy desktop), but why?
We're not talking about just a compiler here. Nobody says "You should try CL.exe, it's really awesome at compiling!" No, they say "Visual Studio is the best C++ IDE you'll ever use." Which is true, but not because of the compiler. And without all the Windows-specific libraries, intellisense and code analysis tools that are second-to-none can't work.
Maintaining two separate development environments, on two operating systems that you have to boot into separately is very annoying. Developing for an operating system other than the one you're running is about as old as the personal computer, I think :).