Yep, we've been working with MS via premier support for the past week, sending gigs worth of text logs. The issue has been escalated to development, but I put CU3 on last night. I'm still seeing the occasional PLE crash but usually it's down to a still-reasonable value, it only floored to zero three times today, and only on one NUMA node. Yet to determine whether that was related to query activity (I have an EE capture running to find that out, but it's late here, and it's Friday... )