The ergonomics for selecting number of GC workers does not take HeapSizePerGCThread into account. This is sub-optimal and can even lead to pre-mature OOME when using a tiny heap on a very large machine, since ZGC reserves 2M of the heap per GC worker.