I read his report, and he is ignoring the peaks, where the servos must make a fast transition. I see peaks of up to 700mA in his data plots, so why he thinks he only needs the torque corresponding to 400mA makes no sense.

I believe the torque ratings on servos are measured at the stall point, so if you need only 21oz-in of torque to hold it static, you'll need more torque than that to move it.