You're gettin into the area of non-linear design here. It is correct that the output impedance of the matched PA does not equal 50 Ohm. The output impedance is cylic, and varies during the period. You can use simulators to define it. And you can use simulators to measure the quality of the filtering involved by just plugging the filter into the simulation of the PA.
At least the filter is matched at one side, this is one good thing. But you might not see a nice 50 Ohm at the other side of the filter if the losses are high, beware of that!
Also, take care you do not put too much power into the SAW. These are acoustic devices with vibrating structures. More power = heavier vibrations = DESTRUCTION! Watch the datasheet.