The reactance of the cables also affects the load sharing. The relative impedance of the individual paralleled cables determines the distribution of current, not just the relative resistance.
Based on my old GE Short Circuit Calculation handbook, the impedance of three, single-conductor 500 kcmil cables in non-magnetic conduit is 0.0276 +j.0373 = 0.0464 ohms/1000 ft. (75C conductor temperature, tinned copper, RHHW, triplexed). Total impedance at 60 Hz is almost twice the resistance. (0.046 versus 0.027)
Cable impedance changes with the installation method. Cables in steel conduit have higher impedance due to the magnetic effect (0.0551 Ohms/1000'). If the cables are routed spaced apart in open air, the impedance is higher because the A phase current's magnetic field is not cancelled out by the B&C phase currents as well as it is when the wires are twisted together in a conduit.
If you parallel cables and bunch all of the A phase cables together, some mutual impedance effects will affect the load sharing. (I measured 20% difference in loading on a 5/phase 1200 amp circuit where all cable lengths were within 1%. The difference was due to the way the cables were laid in the tray.)
Bottom line - keep the cable lengths close to each other (+/-5%) but don't worry about it much. If possible, route the wires in bundles of A,B,C,N to minimize inductive effects. If your load is really 2,000 Amps, throw in an extra 10% to cover unequal sharing of load between paralleled cables.