Most attenuators are not accurate enough to worry about an additional <1% loss either way. If you care about that you probably need a traceable cal, in which case you will have the exact S21 at least for the test fixture used for calibration.
I'd guess that the nominal value of attenuators is generally based strictly on the resistor values, ignoring the additional loss from reflections.
In the real world, the output match of your amplifier is at least as bad so you will have a frequency dependent standing wave with more than 1% ripple anyway.
True, that's all the real math stuff.
But in general, VSWR loss is in addition to attenuation, in terms of rated attenuator.
In other words, if I tell you the DUT (attenuator "A") is -3db with 1.2 VSWR (f=0-6GHz), that -3db is independent of the VSWR factor. I mean it should be, there's no real SWR at 0Hz.
This is why VSWR(f) is important. A spec of "-3db 0-6GHz" is an incomplete spec of the device. I see many time spec sheet that says "-3db VSWR<1.2 0-6GHz". Somewhat better, but we don't know if that means it runs near 1.2 the whole way, or if 1.2 is just a peak in a very narrow band and then runs 1.05 on eitehr side. This is why I like to see VSWR chart before buying.