Here's my attempt to rationalize the rules on EGC sizing:

One of the jobs of the EGC is to clear a fault from an ungrounded conductor on a grounded supply system. In a fault scenario, the EGC is going to be in series with the ungrounded conductor and so their impedances will add. That means that given the available fault current at the source end of a circuit, and given the size and hence trip curve of the OCPD protecting that circuit, there will be a maximum total impedance that is permissible.

Now conductor size determines impedance per unit length, and so length is a clear scale factor here. To continue the discussion we just need to assume some sort of typical or maybe 95% percentile length.

Then the ungrounded conductors will be sized based on the OCPD so that the I^{2}R heating of the conductor doesn't damage its insulation. The EGC doesn't have this concern, as it only carries current during faults, not normal operation. The result is that the EGC can be smaller than the ungrounded conductors. That is most of the impedance "budget" goes to the EGC, since operational concerns already require the ungrounded conductor to be low impedance.

In this picture, an obvious issue is when, for a given OCPD size, the ungrounded conductor(s) have higher than expected impedance. One example of this is when the conductors are upsized for voltage drop. That's a sign that the conductor length is greater than the typical assumed value, which will increase the impedance of both the ungrounded conductors and the EGC. So if you need to bring down the ungrounded conductor impedance by upsizing it, you also need to bring down the EGC impedance by upsizing it.

Another case where ungrounded conductor impedance goes up without a corresponding reduction in OCPD size is the case of parallel conductors and the scenario of one of the parallel ungrounded conductors faulting to one of the EGC conductors. Now we've increased the ungrounded conductor impedance by a factor of 2 or 8 or whatever (not precisely, as ampacity is sublinear in conductor cross section, but same idea). If we also increased the EGC impedance by a similar factor, that would just compound the problem. So we aren't allowed to downsize the EGC relative to a single conductor installation. In fact we should be happy the NEC doesn't require us to increase the EGC size when we decrease the ungrounded conductor size by using parallel sets.

This is a direct effect of a single OCPD protecting our parallel sets jointly. If the NEC allowed us to parallel OCPD, so that each conductor in the parallel set was protected at its ampacity, then it would make perfect sense that each EGC could be sized based on this lower individual OCPD size. But paralleling OCPD is prohibited.

Cheers, Wayne