The datasheet states "Input and output internally matched to 50 Ω", so I've naively connected the input to a matched source and the output to an Agilent spectrum analyzer directly, and the measured results are:
However, I will run some simulations to check if the "not so 50 ohm" matching could cause the difference...