George,
I have not seen anything in print about this but am interested in an answer if you get one. Unlike the GFCI (which looks only for an imbalance in line-neutral current) the AFCI is looking for an arc signature with a threshold amplitude. The longer the circuit and the softer the service the less likely you will get the amplitude needed from a fault. I have experienced failure to trip on AFCIs used on the secondary of small dry transformers (like 5 kva) for local test arrangements. The work that went into developing the devices looked at several hundred in-use receptacles to come up with a 'minimum' available short curcuit current to aim for. I believe that was about 75A.
