I remember every other book on Windows programming saying "Process creation/destruction is expensive, use thread pools (or at least process pools) instead, that's the way to go on Windows". Perhaps this mindset is ingrained for Windows QA team too - they don't have [enough] test cases for such scenarios.
Seems like a feedback loop. It's expensive, so most apps avoid doing that, meaning there's less need to check for performance regressions. Then if there is a regression, it further increases the incentive for developers to avoid it, so in the future even fewer apps will do that, making it even more of an unusual use case, so even less need to test for it..