By reducing the “idle” time of the GCH’s, they are then reallocated faster to “active” transfers. Therefore the efficiency of your Ater PS is higher.
It would be logical to observe such a phenomenom. Just to be sure, why don’t you roll back this timer to the previous value, and check if the traffic decrease ?
A field trial should always end with a roll back to initial value, in order to isolate the effect of the change vs. the natural evoution of network.