Frustrating http connection behaviour in .Net (C#)

Current Status: Red

I seem to be having a bit of a frustrating time with http connections in .Net at the moment.
Basically I am working on a project in which many subsystems use http to communicate with eachother. Once in a while, my .Net client will fail with a "System.Net.WebException: The underlying connection was closed: Unable to connect to the remote server." exception.
The remote server can be an IIS server or a Jetty server running on Linux - I see the same problem with both.

To try and reproduce this problem I wrote a very simple client app that requests a single static page from a web server. It loops round making the request repeatedly, up to 100000 times.
Here is the static Main method from my console application:

[STAThread]

static void Main(string[] args)

{

    int i = 0;

    try

    {

        for (i = 0; i < 10000; i++)

        {



            HttpWebRequest req = (HttpWebRequest)

                WebRequest.Create("http://proxima/test.html");

            using (HttpWebResponse resp =

(HttpWebResponse)req.GetResponse())

            {

                byte[] buffer = new byte[1024*1024];

                int read = 0;

                Stream stream = resp.GetResponseStream();

                do

                {

                    read = stream.Read(buffer, 0, 1024*1024);

                }

                while (read !=0);

                stream.Close();

                resp.Close();

                req = null;

            }



            if (i % 1000 == 0)

                Console.WriteLine(i);

        }

        Console.WriteLine(i);

    }

    catch (Exception ex)

    {

        Console.WriteLine("Exception at iteration " + i);

        Console.WriteLine( ex.ToString());

    }



}

For this example I created a very small html file to put on my server - only 50 bytes.
I see some extremely peculiar behaviour when I run this application; generally the first time it runs, it will process many thousands of iterations before falling over with the aforementioned exception. Once it has failed, if I attempt to run it again straight away the application fails in a matter of a few 10's or 100's of iterations. Why?
It makes no difference to the overall behaviour if I swap my HttpWebRequest stuff for a System.Net.WebClient.
It makes no difference if I insert a GC.Collect() in the loop.
What is going on here and what is 'remembering' the fact that the client has previously failed?

I have to admit to being stumped for the time being.
Does anyone else see this behaviour?