In the first model all the mass is on one side of the X axis, so all the mass is going in the same direction with respect to rotation about x. Whereas in the second case half of the mass is on each side of the X axis. So half of the mass is going positive rotation about X and the other half negative rotation so the cancellation.
In the beginning of troubleshooting it helps to not assume one code is right and the other wrong. First, compare the theory between the two to make sure that they are computing the same thing. It could be that there is something defined slightly different, but equally valid approaches, between the two codes. Have you compared the actual implementation?
p.s. sory if I am repeating any of the prior discussion...I started replying to this post when there were no replies. Got sidetracked. Then finished and when I posted then I was able to see that whole discussion (page refresh).