BenchmarkRandom.java

package net.moznion;

import me.geso.nanobench.Benchmark;
import me.geso.nanobench.Benchmark.Bench;

import java.util.Random;
import java.util.concurrent.ThreadLocalRandom;

public class BenchmarkRandom {
	public static class RandomNumberGenerator extends Thread {
		private static final Random RANDOM = new Random();

		@Override
		public void run() {
			RANDOM.nextDouble();
		}
	}

	public static class ThreadLocalRandomNumberGenerator extends Thread {
		@Override
		public void run() {
			ThreadLocalRandom.current().nextDouble();
		}
	}

	@Bench
	public void withRandom() throws InstantiationException, IllegalAccessException {
		runThreads(RandomNumberGenerator.class);
	}

	@Bench
	public void withThreadLocalRandom() throws InstantiationException, IllegalAccessException {
		runThreads(ThreadLocalRandomNumberGenerator.class);
	}

	private <E extends Thread> void runThreads(Class<E> threadClass)
			throws InstantiationException, IllegalAccessException {
		for (int i = 0; i < 1000; i++) {
			threadClass.newInstance().start();
		}
	}

	public static void main(String[] args) throws Exception {
		new Benchmark(new App()).warmup(10).run(1000).timethese().cmpthese();
	}
}

Score:

withRandom: 52 wallclock secs (18.70 usr + 24.70 sys = 43.39 CPU) @ 23.05/s (n=1000)
withThreadLocalRandom: 52 wallclock secs (18.27 usr + 24.81 sys = 43.08 CPU) @ 23.21/s (n=1000)

Comparison chart:

                           Rate  withRandom  withThreadLocalRandom
             withRandom  23.0/s          --                    -1%
  withThreadLocalRandom  23.2/s          1%                     --