One more interesting detail... I tried this on SQL 2008, and it works as you described, but with a twist. Each time the procedure runs, it produces a single row result, but the row produced is different each time. While the result appears to be random, after running this many times, it seems that the row selected is actually one of a small set of possible results (though that may be that I didn't run it enough to get the full effect of randomness).

Too weird.

😉