I can't make any reasonable statement about the likelihood of Yahoo having done this, but it seems easily possible to do in a few minutes. Forget hours, unless there is some fundamental aspect of the data structure used that I'm missing.
For example, I have a nightly process (plain jane RBDMS) that searches over 85 million records to match up with 100K records where the matches may be exact or provide potential matches from the larger table back to the smaller. This process typically takes less than 20 minutes and the system runs on what I would define as relatively average hardware.
I am assuming that the user accounts are logged in some way that keeps them from having to do some awful full text scan of blog comments for each of the 20 million users checked.
[EDIT: Rechecked log files, runtime is 15-20 mins instead of 5]