I have experienced insufficient sampling for high energy states in particular windows. The spring constant is not enough stiff to keep the value of the reaction coordinate. I know that you combine windows where the harmonic potential used differs since you provide the set of k values im wham however I am wondering if it could be better to sample those windows by increasing the spring stiffness or by reducing the spacing interval for the reaction coordinate.