Chances are that because, in your test, you only call this once, the delay you see is the C# code being JIT compiled.
A better test would be to call this twice, and discard the timings from the first time and see if they are better.
An even better test would be to discard the first set of timings, and then run this many times and take an average, although for a very loose "indicative" view, this is probably not necessary.
As an aside, for this sort of timing, you are better off using the System.Diagnostics.Stopwatch
class over System.DateTime
.
[EDIT] Also, noting Mant101's suggestion about proxies, if the setting no proxy fails to resolve things, you may wish to set up Fiddler and set your request to use Fiddler as its proxy. This would allow you to intercept the actual http calls so you can get a better breakdown of the http call timings themselves from outside the framework.