I think there is an important difference between "saturation" and loss of phase coherence by a gradient pulse. A gradient pulse only dephases spacially. So at any arbitrary point in a sample, nearby nuclei would still have phase coherence with each other (gradually becoming less coherent as the distance between them increases along the Z axis). A truly saturated sample has lost all phase coherence everywhere. Also, the loss of phase coherence by a gradient can easily be reversed, again suggesting that this is different from true saturation which can not be reversed (without restoring M_z first).