I run the Surface benchmark with -resolution 400. When I start moving the mouse over the scene fps noticeably drop. One factor is time spent in TriangleMesh.computeIntersectsFace(). Two reasons are obvious:
1. TriangleMesh.impl_computeIntersects() doesn't break from the loop when an intersecting face is found
2. TriangleMesh.computeIntersectsFace() itself generates a lot of short lived objects (Point3D), which causes significant delays due to GC

I reproduce the problem on Windows and it kills performance on embedded.