I am having trouble coming up with a scenario where this would actually make sense -- you can always delegate specific permissions to a named user to get them the specific admin-style rights they need.
As for the question at hand, direct risk isn't any greater than any other web application inasmuch as a web app is a big honking hole through your firewall. The indirect risk is very, very scary. You are trying to turn the clock back to 2000 when IIS5 was setup to run as local system making every single case of "IIS can be made to run arbitrary commands" into "anyone can own your box over port 80."
If you do have to do this, I'd consider putting firewalls behind the server too. That way, when it does get rooted, you've got some defenses. I'd also use unique accounts, etc.