Mathematically this "string theory" (sorry, couldn't help myself) is saying the following:
sqrt[ (l/2)^2 + b^2 ] < min(x,y)
where l = length of box, b = base of box, and x and y are the respective widths of the two hallways.
However, this ignores the fact that the pivot point on the box around the corner can slide as you push the box around the corner. So the pivot point could essentially slide from one end of the box to the other (at some point sliding through the midpoint you are measuring for), which allows a lot more leeway for the box's acceptable dimensions.
I worked on this problem a bit this morning, but I don't have my trusty TI-89 calculator to help out. This is actually a calculus problem that I have a difficult time explaining without posting some sketches. Maybe I'll do that in the near future.
You speak with knowledge, worthy adversary, but with my HP-48g, I would defeat you! </bad overdub due to RPN>
Sometimes you can get around it by leaning the box at a vertical angle, thus reducing the length that actually needs to get around the corner.
If you don't know the size of the box in advance, hold the string at the corner and make the string the longest possible length that can sweep this arc. Take the sting with you to the box.
What you are really doing is measuring the distance to the corner from the closest wall and then making sure the line from the box's midpoint to one of its corners is not longer. There are edge cases (ha!), but this generally works.