Ok, here's my thought.
Suppose I want to measure the NF at 2 MHz. I can use an external mixer with say 100MHZ LO and get an SSB measurement, say the USB (102MHz).
The measurement bandwidth of the 8970B is fixed at 4MHz (don't know if it's higher), so the instrument will measure noise power from 100MHz-104MHz. Provided that there's no ideal mixer and a LO leakage will always exist, the measured power will be higher, creating a false measurement.
I don't know if this situation can be overcome by the calibration. I mean, does the calibration+measurement cancel out the lo leakage contribution?