Hi, thanks for sharing this issue.
At the moment we aren't aware of any memory leaks in NCrunch itself that could cause this kind of behaviour, but when there are 8000+ tests involved, sooner or later this is probably going to happen.
It's not an uncommon problem either. It's not easy to verify cleanup operations for each test, and leaks can be quietly introduced without anyone being aware of it until much later down the line.
A while ago we added a feature that more or less negates this problem. Set the
test process memory limit in your NCrunch configuration to something sensible. If a test run completes and the runner EXE is holding more memory than the limit, it will be automatically recycled and the problem is gone. Unless you're leaking memory like a garden sprinkler, you likely won't notice any loss in performance as the recycling of the processes will be infrequent enough to not have an impact on your test cycle times.
