为什么Parallel类的For方法会随着时间的推移变得更快？

Question

我正在做一些测试，以找出生成的UUID编号中的某些重复项，并且我使用不同的方法来执行计算。它们是multithreading，normal for loop和for method of the Parallel class。 Multithreading是明显的赢家，normal for loop则更宽松，Parallel.For方法随着时间的推移变得越来越快。我用Parallel.For做了10次测试，它从18.08秒减少到11.80秒。我一直在启动项目，一次执行一个测试，所以我相信它不能存储在内存中。在测试期间，我没有在计算机上做任何其他事情。

这些是Parallel.For方法的测试结果：

1: 00:00:18.08
2: 00:00:17.04
3: 00:00:16.79
4: 00:00:15.15
5: 00:00:13.00
6: 00:00:12.69
7: 00:00:12.66
8: 00:00:12.54
9: 00:00:12.33
10: 00:00:11.80

对此有逻辑解释吗？

如果需要的话，这里是代码：

// In Main
Parallel.For(0, 10, i =>
            {
                TestForUUIDDuplicates(i + 1);
            });

static void TestForUUIDDuplicates(object loopIndex)
    {
        if (!(loopIndex is int))
        {
            throw new InvalidCastException();
        }

        var uuids = new List<string>();

        for (int i = 0; i < 1000000; i++)
        {
            if (i % 100000 == 0) System.Console.WriteLine("Reached {0} in loop {1}", i, loopIndex);

            uuids.Add(GenerateUUID());
        }

        System.Console.WriteLine("Finished adding UUIDs in loop {0}.", loopIndex);

        var duplicates = uuids
            .GroupBy(uuid => uuid)
            .SelectMany(group => group.Skip(1)).ToList();

        System.Console.WriteLine("Finished finding duplicates in loop {0}.", loopIndex);

        foreach (var duplicate in duplicates)
        {
            Console.WriteLine(duplicate);
        }

        if (duplicates.Count == 0)
        {
            Console.WriteLine("No duplicates found in loop {0}.", loopIndex);
        }
        else
        {
            _duplicates.AddRange(duplicates);
        }

        Console.WriteLine("Finished loop {0}", loopIndex);
    }

    private static string GenerateUUID()
    {
        // var ticks = DateTime.Now.Ticks;
        var ticks = DateTime.Now.Ticks;
        var guid = Guid.NewGuid().ToString();
        return ticks.ToString() + '-' + guid;
    }