Weird Behavior with New Geometry Optimizer

I recently tried using Q-Chem 6.0 to perform the geometry optimization of a complex. In particular, I began with a distorted geometry of the transition state and optimized towards the reactants as I was curious which specific reactant conformation is connected to this transition state. When I used the old geometry optimizer, everything works smoothly (see attached P-Pyridone_TS_Reverse_old_opt.out). However when I used the new geometry optimizer, the initial geometry optimization step seems totally crazy, especially since the exact Hessian is read in. The geometry optimization ultimately reaches a totally different local minimum (see P-Pyridone_TS_Reverse_3.out).

Is there a way to prevent such a crazy initial geometry optimization step?

Sample Output Files