I don't know the details but if you're too close (electrically/conductor length, not physical distance) to the transformer then something about the transformer interferes with the measurement and leads to a false reading.
Thinking about that and looking at, say, the diagram in the IET design guide here:
In the event of a "three phase fault" as point (a) with a symmetric/balanced system the centre of the three line short would be a virtual neutral point due to all the currents balancing out, hence the PEN would have zero current and so the PEN impedance would have zero effect.
So the true worst-case PFC is computed from the assumption of a zero-impedance link between, say, L1 at point (a) in the diagram and the transformer star centre/ true Earth:
PFC0 = Uoc / |Zx + Zd|
However, the common thing to do is measure the PSSC L-N and double it, this would be:
PFC1 = 2 * Uoc / |Zx + Zd + Zpen|
Normally the
assumption is that neutral and line conductors are the same impedance (i.e. Zpen = Zd) so we have:
PFC1 = Uoc / |0.5*Zx + Zd|
So PFC1 is under-estimating the effects of the transformer source impedance Zx in this simplified model of a three-phase transformer as 3 independent Thévenin equivalent sources and giving a higher fault current than the true PFC0 value.
But as you mentioned you can measure the line-line PFC and this is potentially more accurate as then you are measuring:
PFC2 = sqrt(3) * Uoc / (2 * |Zx + Zd|)
Rewriting this as PFC2 = (sqrt(3) / 2) * (Uoc / |Zx + Zd|) = 0.866 * PFC0
Hence PFC0 = 1.1547 * PFC2
In this case we don't under-estimate the transformer impedance, nor do we make the assumption that the PEN/neutral path is identical to the line conductors. As such this is probably the best way to measure the worst-case three phase fault current is to do line-line and multiply by 2/sqrt(3) = 1.1547