Then why have usernames (displayed) at all? Why is a picture so different than a username in terms of degrading the painting of a "portrait" of ideas? You could drop them altogether or replace them with a random string (generated each page view).
Imagine watching the NBA on TV with no names on the jerseys and all the faces blurred out. Writing style without names would be (in some but not all cases) like trying to identify players by their shoes, tattoos or morphology (read: mostly freakish tallness).
That it would be unpleasant to the watcher if the names and faces were blurred out, implies that the celebrity nature of the activity holds much importance.
Haha, just saying, FWIW, because your comment was interesting.
How this relates to the present topic I don't know; hell, I'm not even sure this hypothesis is correct. I just think it's interesting.
I would posit that the celebrity nature of professional sports is the biggest contribution to making them becoming huge money-making enterprises in the last few decades. If you made all players anonymous, I would bet money that the enterprise would shrink and become a more local phenomenon.
Two, getting a username is considered a sunk cost, one many are willing to pay, but setting up a picture is an additional cost; some will pay, but I would wager many of us here won't bother. However, once you set up the option, there will be a "ah, they're the ones who like to update their profile pic!" and "ah, they're among those who simply don't care!" Assume many people don't bother, you still need to make do, which is your current situation.
Third, user-selected pictures carry a lot more information than just the usernames, and this information may simply detract from the core of the site, which is the content and discussion. If you treat usernames just as differentiating handles, they currently serve exactly that purpose.
Of course, you will be right to say that pictographs can be more efficiently retrieved from memory here, but the case doesn't seem strong enough to spur investing time into accommodating the functions in the code. I agree with pg above in this thread.