As far as memory usage goes these benchmarks are completely irrelevant. They use very little memory and they measure it in a way that is unsuitable for a memory benchmark in the first place.

The problem with tracing garbage collection is that it requires a lot (~50%) of spare memory at all times to give the GC time to catch up without slowing the program down too much.

This is the kind of argument that used to be trotted out all the time for Java and Perl before it. If we always went by what was offered today, we'd never use anything new.