The Perez diffuse sky model produces strange results. It seems to be related to the epsilon calculation for lower zenith angles.  