Recently I've seen some C# projects that use a double-checked-lock pattern on a Dictionary. Something like this:

private static readonly object _lock = new object();
private static volatile IDictionary<string, object> _cache = 
    new Dictionary<string, object>();

public static object Create(string key)
{
    object val;
    if (!_cache.TryGetValue(key, out val))
    {
        lock (_lock)
        {
            if (!_cache.TryGetValue(key, out val))
            {
                val = new object(); // factory construction based on key here.
                _cache.Add(key, val);
            }
        }
    }
    return val;
}

This code is incorrect, since the Dictionary can be "growing" the collection in _cache.Add() while _cache.TryGetValue (outside the lock) is iterating over the collection. It might be extremely unlikely in many situations, but is still wrong.

Is there a simple program to demonstrate that this code fails?

Does it make sense to incorporate this into a unit test? And if so, how?