The scenario @Tigga described ^ is exactly what I see. Here's a graph of my CPU (top) and heap + Global Age (bottom) for the last week or so. You can see that my CPU usage drops steadily the longer my global has been alive. I believe this is because some small leak slowly changes the heap characteristics such that GC transitions from almost always happening during my user code (incurring massive CPU cost) to almost never happening during my code. I know it's a difficult problem, and I certainly don't have a solution, but I dream of not having to deal with the volatility of GC. I'm confident that the immortal
Game object changes will help immensely.