I've been working on a project that provides an alternative to cdn.d.n. It
is based on http redirections.
An introduction can be found on the site itself[1], and a comparison to
cdn.d.n at [2].
I now have setup a test instance in a host in Canada to allow people to
actually test and use the system. This host is http.debian.net.
This is very bad idea, instead of having auto mirror selection mechanism you've implemented classical Single Point of Failure
Proper implementation should loop over existing mirrors, starting with less latency/faster transfer and then fallback/load balance to another,
if file is not available. Unfortunately your solution is not solution at all. Something close to proper implementation is used for fedora infrastructure.
3xx per file redirect is not a scale-able solution.