It's bad if users are trained to only recognize _your_ phishing attempts :-)
I'm not sure I understand which users jacques_chester is talking about.
There are users that can recognize phishing, and they are entitled not to care about your teaching. And then there are those that can't recognize phishing - or perhaps don't even know about it - but I'm pretty sure any user would start caring when they find out someone else can gain access to their email/bank/facebook/whatever online service they use if they aren't careful.
To avoid training users into thinking it's another drill, perhaps it's a good idea to 'attack' them at random intervals, and wait a few months before repeating (thus giving you enough time to prepare the new attack; giving the users enough time to forget about the threat, and to account for new arrivals).
I'd rather be embarrassed by the local BOFH, rather than be a real victim
Now that's the best idea I've heard all morning. You should be running Oxford's IT dept!