Complicated question regarding a protocol to decrease the transition

We are trying to calculate the energetic cost of the transition of two structures of TFIIB protein. I am using the CNS script available on the molmovdb to obtain multiples states along the transition between both structures. I am doing this from the xray to nmr structure on both ways. My problem is the following, after a few attempts I found a protocol to decrease the hysteresis of the transition. I begin with 500 frames and every frame energy minimizated with 500 powell steps. I tried to decreased even more the hysteresis and increase to 1000 steps with the same 500 powell steps. My problem began when in other attempt to decrease the hysterisis and I made 1500 frames with the same 500 powell steps, as you can see on the graph that I attached, It appears that the hysteresis increased with the 1500 steps. When I repeated the 1500 frames and minimizated with 1000 powell steps, the hysteresis decrease.

Do you think that this behavior is correct? Does it look strange that I have any energetic barrier of the transition?

I would be very cautious about assuming that a linear interpolation between two structures represents the thermodynamically most probable trajectory of motion. I don't know why using more frames would increase hysteresis -- presumably you mean that an energy difference resulted when the protein nearly finished its morph trajectory. Maybe the coarser interpolation jumped over a barrier and found a more favorable path. If you clarify what you are doing I may be able to provide a hint, though I think most likely I do not have a rigorous answer for you.