Currently G1 determines the number of parallel worker threads during young gc only on total heap capacity.

This is very inaccurate in situations when G1 is used to achieve small pause times on large(r) heaps (e.g. data heavy workloads like https://jet-start.sh/blog/2020/06/09/jdk-gc-benchmarks-part1) this is wrong and detrimental to performance.

Investigate a better policy for this.