Hi All
Just an update,
I got a test board made up at OSH Park with the correct 0.2 prepreg. The results were better, but still not acceptable.
I then realised I had done a silly thing. I had extended the 0.2mm ground plane under the SMA connector center pin pad. This area yields about 0.9 pf. A capacitance this size at 3 GHz will certainly bring the impedance way down and be highly reactive. I modeled the layout with and without the pads and voila, correct answer. With the pads, I get a virtually identical S11 response as the OSH Park board. Without the pads, I get the impedance you would expect from the theory.
I modified the board design with keep-outs under the SMA center pin pad.
I am 99% certain that will fix it.
Just an aside I soldered two of the SMA connectors back-to-back and got a very good response.
What a knucklehead!!
That was God punishing me for not using the brain.