I wonder if they implemented a copy-on-write fork syscall in NT. Otherwise it will be slower (and use a lot more memory) in some scenarios.
Edit: And cow fork only makes sense if there is memory over-commit. So to be fully featured it would need a separate memory subsystem with memory over-commit.
No, you don't need overcommit for COW fork. You can fully commit the writable pages of the new process. Yes, that uses a bit of pagefile space, but disk space is cheap, and this approach works fine. You can (and should, if you care about robustness) turn off overcommit on Linux too.
Edit: And cow fork only makes sense if there is memory over-commit. So to be fully featured it would need a separate memory subsystem with memory over-commit.