0.05% ration accuracy should be well good enough. A new calibration should be able to compensate for most ratios. There may be a few ratios that may not be fully compensated with a calibration. This are those effecting the slow slope of the MS ADC and to a small part the symmetry of the pos and neg reference. The slow slope is at 1/128 the normal reference and thus a special ratio, so there may be no correction. Anyway it uses an external resistor too, so this part should not be that critical and even 0.5 % may be OK. The symmetry also enters a little and not sure if there is a correction - if not used for the slow slope, I would not expect if the the symmetry either. The importance / required accuracy should be about comparable to the slow slope part.
The amplifier gain part is very likely fully corrected - so no extreme needs there.
The absolute gain for the reference is definitely corrected / part of the calibration, but the resistors are together with the symmetry anyway.
From what I have heard the failure may effect only some resistors (e.g. failure of bonds). So one may get away with only a partial replacement.
For the TC, it is not so much about the absolute TC, but the TC matching. The TC matching in the array can be very good (e.g. 1 ppm/K range).
For separate resistors it can help to use with resistors from the same series/construction, as there is a good chance that the square part of the TC is relatively fixed and mainly the linear part varies. So the matching between 2 resistors at 5 ppm/K has a chance to be better than 5 ppm/K.
It also help to get the best TC grade, as the lower grades may have the best ones selected to go to the better grades.
The resistor for the integrator input can effect the linearity. So this resistor should be low TC (preferably better than 10 ppm/K) for good performance. With not so good thermal couling here the abloute TC can matter, not just the relative. Many Keithley DMMs use rather frequent measurement of the ADC gain in the background and thus may not need very good TC matching for the other resistors. However I don't know for the K2000/K2015.