equal to *gc-trigger*, and that was causing GC not to happen in some cases on x86. He said changing > to >= in sub-gc fixes this problem.