Is there any code you have running under test that might trigger the compatibility mode assistant? (i.e. are you interacting with any deeper level or legacy APIs?)

From a couple of the other issues you've reported earlier, I'm wondering if for some reason your system is having issues with the hosted task runners. There's a configuration setting that you may want to try to see if this helps. If nothing else, this may be useful in order to narrow down the source of instability. Try setting the Use hosted task runner global configuration setting to 'False'. You'll notice that the NCrunch task runners will run under different names, but there should be no functional difference.