It's nothing new to me and probably nothing new to most folks who work with MySQL on a regular basis that the username alone does not identify a MySQL user, but a combination of username@host.
But if you forget about it for a moment, you can easily begin to wonder about special behaviors. That just happened to me, so I just thought that I have discovered a very evil bug - but looking twice, it became quite clear to me why MySQL behaved this way.
I have two computers, both with MySQL servers, both with users called mpopp - but each server has two of them - one is mpopp@localhost and the other is email@example.com__ (both with the same permissions), so it allows me to access user mpopp from each of my PCs. I created a little stored procedure on one computer and then copied this procedure to the other one - together with the new DEFINER option that's new in MySQL 5.0.20. But I didn't think about it (and didn't look carefully) and created the new procedure with the option DEFINER=mpopp@'10.0.0.1__' (while being logged in as user mpopp@localhost).
It was quite a shock when I called "SHOW CREATE PROCEDURE" and there was no output for the definition - so it looked like my procedure was gone. Also in information_schema.ROUTINES I couldn't find the definition, although the procedure itself was listed.
It became clear when I logged in as root and finally found that the procedure definition was still there. I looked a bit more carefully and found the mpopp@'10.0.0.1__' user in the DEFINER column. Of course, you can only output the procedure definition from the user that created the procedure (or from a root user, obviously), and in that case, it was user mpopp@'10.0.0.1__' and not mpopp@localhost (even though this was the user that I was logged in when I created it).
That makes me think - wouldn't it be more intuitive if the user name alone would identify a user and the host definition would be separated by a 1 : n relation? So for example - there's a user mpopp with all the permissions and there's a separate table that does nothing more than to define from which hosts the user has access? I know, to change that would break backwards compatibility completely - but on the other hand, it might be less confusing. I would be curious what other people think about this.