The serial part to free the collection set on huge heaps with a very large collection set is very slow.
E.g.
Free collection Set: 430.7ms
Free collection set serial: 421.2ms
Young Free collection set: ... Avg. 4.2 ... Workers: 213
Non-young free collection set: skipped
Heap size: 1.5TB (49152 total regions), eden size: 11793->0, survivor size: 5007->5007, old 871 -> 871, humongous: 16 -> 16
The problem is that inserting free regions in the free list without any ordering takes very long due to being and O(n^2) operation.