As detected by Bellaz89 in #5776, the GUI page global object is incorrectly rooted into the simulation context.
In my opinion, there is a tiny, tiny, microscopic chance, that this could cause the GC segfault in SpiderMonkey. In that hypothesis, the upgrade to 45 made the JS engine less permissive about this mistake. I can't reproduce the segfault, so I can't make sure.
In any case, this should be fixed.