Below 1nH I would go for an accuracy no better than +/-30% (maybe worst).
Inductance measurement in a PNA is an indirect measurement, calculated from an impedance measurement.
You can use a VNA to measure inductance. You just need to measure at a high enough frequency so 1 nH presents a impedance near 50 ohms. That also means you need a good calibration. You can use either 1 port or 2 port measurements, preferably both.
The 4287A is just a 1-port VNA anyway. The bridge type LCR meters are probably too low in frequency to measure 1 nH, however they are very accurate for certain ranges.
You don't have to measure the inductance directly. You can resonant it with a known capacitance and measure the suck out. There are methods using transmission lines too. One method may be better a finding the Q, another better for L or SRF. You need to research the different methods.
In this case, I would prefer to measure L and Q over wide frequency band. Hence using LC oscillator is not suitable for my application.
But madengr gives a very good idea: Is it if |Zin| near to 50ohm, measurement accuracy can be greatly improve, assume inductor are measured using 2ports VNA?