For sensitivity measurements in general lots of attenuation is needed, so one can assume enough isolation is provided to separate the input and output from VSWR interaction.
A 50-75 transition will generate a nominal 0.2 dB mismatch loss. In total 0.4 dB total mismatch loss assuming both ends are 50 ohms and the attenuator has a perfect 75 ohm impedance. This is also the value at the end of rfeecs his table. This offset could be corrected for.
However it could just be left uncorrected as well. Its effect then could be incorporated in the measurement uncertainty.
To be really correct there are some additional steps to be taken, but assuming a system where both the source and load return loss are > 15 dB in 50 ohm and the attenuator is 20 dB in 75 ohm, together with u-shaped distributions of the reflections, it is also boils down to about 0.4 dB at 95% confidence. (no wonder).
Assuming all other errors add up to say 1 dB uncertainty, the combined uncertainty will be 1.1 dB. So nothing to worry about.