Op的答案是迄今为止提出的和建议的解决方案中***的.这是我机器上的时间:

Op's answer is the best out of the proposed and suggested solutions so far. Here are timings on my machine:

public static class Class1
{
    // 21ms on my machine
    public static List<T> FindAndRemove<T>(this List<T> lst, Predicate<T> match)
    {
        List<T> ret = lst.FindAll(match);
        lst.RemoveAll(match);
        return ret;
    }

    // 538ms on my machine
    public static List<T> MimoAnswer<T>(this List<T> lst, Predicate<T> match)
    {
        var ret = new List<T>();
        int i = 0;
        while (i < lst.Count)
        {
            T t = lst[i];
            if (!match(t))
            {
                i++;
            }
            else
            {
                lst.RemoveAt(i);
                ret.Add(t);
            }
        }
        return ret;
    }

    // 40ms on my machine
    public static IEnumerable<T> GuvanteSuggestion<T>(this IList<T> list, Func<T, bool> predicate)
    {
        var removals = new List<Action>();

        foreach (T item in list.Where(predicate))
        {
            T copy = item;
            yield return copy;
            removals.Add(() => list.Remove(copy));
        }

        // this hides the cost of processing though the work is still expensive
        Task.Factory.StartNew(() => Parallel.ForEach(removals, remove => remove()));
    }
}

[TestFixture]
public class Tester : PerformanceTester
{
    [Test]
    public void Test()
    {
        List<int> ints = Enumerable.Range(1, 100000).ToList();
        IEnumerable<int> enumerable = ints.GuvanteSuggestion(i => i % 2 == 0);
        Assert.That(enumerable.Count(), Is.EqualTo(50000));
    }
}