This has showed me that I may need to change my testing framework. Rook and Queen outposts showed up as being very good in self play (around 10-15 elo). But seeing the results here I tested two versions with and without outposts against a common opponent (SmarThink 1.6). Against SmarThink Outposts are a pretty strong detriment (-35 +- 12 elo).
This seems to mean that it will be very tricky for me to test scalability for kingsafety changes, which I believe is one of the places where Nirvana could use quite a bit of work.