For ergonomics decisions (young gen size) we need a fairly accurate measure of how the various sources of heap roots contribute to garbage collection time.

With JDK-8213108 the method to distribute card scanning costs radically changed as scanning the heap roots is a single pass. It distributed the total time according to the number of cards the various roots contribute.

Since the hot card cache is small and either completely unused (not doing lots of refinement) or if there is lots of refinement, the cards in there are dwarfed by the other sources of heap roots, the change did not incorporate this count in the calculation.

Fix this and investigate impact.