It comes down to whether the Narda switch and the cables/adapters used
to connect it are of a higher expected quality than whatever you are
trying to measure. Are they? At HF, it would not usually make sense
to worry about de-embedding microwave-grade interconnects. At higher
frequencies closer to the accessories' rated limits, it can be a
bigger deal.
By definition, the measurement's reference plane is defined by where
you placed the standards when you did the calibration. In general,
the farther these components depart from an ideal 50-ohm transmission
line at the frequencies of interest, the less likely that you can move
the measurement's reference plane to the downstream side simply by
calibrating after the additional components. At some point, you will
need to de-embed the accessory hardware by characterizing it
separately, in order to remove its influence from the final
measurement data. It's hard to say exactly when/where this is
necessary.
For instance, when you take a wideband sweep at a remote reference
plane, you are relying on your analyzer to interpolate its error-
correction terms between discrete calibrated points. Often, cables
and accessories will exhibit a lot of VSWR ripple that the calibration
process will 'undersample.' Calibration alone can't take the
accessories out in this case, but a good deembedding algorithm can.
If you are making a narrowband measurement, or a measurement at a
single frequency, you will probably be fine with calibration alone.
Joel Dunsmore, who has worked on VNA stuff at Agilent since the 8753As
in the mid-1980s, just came out with a nifty book on the subject:
http://www.amazon.com/Handbook-Microwave-Component-Measurements-Techniques/dp/1119979552
But there are still no "Yes, you can do that," or "No, don't do that,"
answers to be had, in the general case.
-- john