This is my hacky approach to the problem. This runs the GC much less than expected but more than what would be acceptable (@chergert's idea make it runs much less than my original approach). But it definitely holds the GNOME Shell memory at the same plateu for a long, long time, helping gnome-shell#64 (closed) not happen.

Some ideas that are worth considering:

• Use a timeout instead of an idle add (it appearently didn't have the same effectiveness though...)
• Don't force GC but instead improve the current heuristics
• What else?

Issue: #140 (closed)