The real bug is in the matrices blending function which doesn't handle non-uniform scaling correctly. I've minimized the occurence of the bug by calling the blending function only when the influence is smaller than one (woah, optimisation AND bug fix!) This should make the bug disappear approximatly 90% of the time since people don't use influence all that often (also, this only applies to constraint that are alone, not using influence IPOs to switch between two constraints). I'd solve the blending function bug, but I haven't had much time to dig into it really. Incidently, by reverting to the previous code, this solves bug #1069 http://projects.blender.org/tracker/index.php?func=detail&aid=1069&group_id=9&atid=125