> I don't think it's even close. Benchmark game is one (biased) reference.
I agree.
1. Benchmarks tend to do one thing, then repeat that one thing 25000 times. Non-server programs don't do that[1] so you don't get to amortise the time spent on JITting a code snippet across 25000 executions.
2. The runtime basically requires exclusive use of a core for short periods (GC, JIT). If you're running on a system that costs per core, you're using (worst case scenario) an extra core for each Java program. The benchmarks happily give multiple cores even to sequential (non-parallel) benchmarks.
[1] Even many server programs don't do that.
