In theory, performing a calibration at a port at the far end of any length of an ideal arbitrary line connected to an ideal VNA gives you the full VNA accuracy for measurements at the far port.
In practice, the deviations of both from non-ideality causes problems.
A real VNA will have a finite dynamic range. Line loss will eat into that dynamic range by twice the line loss, giving you noisier readings. The noise floor can be clawed back somewhat by taking many averages, for both cal and measurement.
A real transmission line will be unstable in length (with temperature and flexion and strain) and in impedance (tempco of dielectric and conductor dimensions). Any change of electrical parameters between calibration and measurement will give you a systematic error in the measurements, which cannot be averaged out. In the worst case that the length changes by half a wavelength, having VNA calibration would give you a worse 'corrected' measurement than just normalising for feeder loss and nothing else.
Even if the line stays totally stable, phase/frequency noise in the VNA source will also change the phase shift of the line. This is not usually a problem for most measurements, but as you say this is a 'long' feeder, source phase noise could become relevant.
Make an S11 measurement on the open feeder and observe how the phase of the reflection changes, at trace to trace timescales that can be averaged, and hour to hour ones that can't. You might want to bend and straighten the cable to see the effect of flexion, if you're going to move it during measurement. Then decide whether you have the stability to meet your measurement accuracy requirements.
As to whether multiples of some fraction of a wavelength will be better or worse ... For any reasonable 'long' length of cable, for any reasonable measurement bandwidth, the length of the cable will whizz round the Smith Chart many times, so you don't get to choose. If the cable and the VNA are reasonable approximations to 50 ohms to start with, then the effect of cascading them will not be too severe, in terms of resonances and suck-outs. Any that do occur will simply exacerbate the problem of working over a long unstable cable - suck-outs will still degrade SNR by twice their depth, 'filter-like' resonances will further amplify the effects of source phase noise and cable-length instability issues.