[modeller_usage] Strange results from loop modelling

I have a dimer in which I have used Modeller to fill in a missing loop region (one in each chain). I started to fill in the two gaps at the same time with the automodel class. Then I selected the best structure and I did loop refinement with the loopmodel class but only on one of the loops. From this step I took out the 10 best structures and did loop refinement on the other loop. Now the strange thing is that it seems that Modeller has refined the second loop in exactly the same way for all the 10 starting structures. The loop from the first loop is different but each refinement of the second loop is identical.

I have not been able to find any error in the log-files or in my script and Modeller has not complained about anything. So my question is: are these results reliable? Is it possible that the loop modelling behaves like this?