I've now dissussed privately with marcomdd about this question.
Effectively the most formal method to know the noise temperature introduced by an in line 2 port network (sma f-f, in this case) pass trough the knowing of the Available Gain (Gav).
The Dissipative Attenuation defined on "Microwave Attenuation Measurement - Warner - IEE Press" is:
dissipative attenuation=|s21|^2/(1-|s11|^2)
and is an old definition, may be valid if Gammasource=0.
In any case, the spar measurement of a low loss, coax DUT depend strongly on the pin recession of both the the couple of the ports (DUT and VNA).
Often you may assume that the VNA port is completely corrected, but it's not true. The pin of a 3.5mm connector may be recessed from 0 to 3mils.
The error introduced by a 3 mils recessed may be important and erroneously considerated as part of the DUT.