Sounds like an unloaded 2000kVA transformer would be a large low power factor load all right.
Low power factor but not a large load. I've just recently had a quotation in for a 950kVA transformer.
This from the proposed manufacturer's spec:
Total rating: 950kVA
Vector group: Ddoyn11
No load loss: 2550 watts
And the connection surge could be hard on the generator.
I think the OP is wanting to test a range of transformers.
A generator large enough to power the largest transformer should do bearing in mind that the generator power factor may be 0.8.
Maybe inrush could be a problem but if it's for test purposes only the generator voltage could be raised after the transformer is connected. I don't know if all AVRs have that facility. The ones we make (not mainstream business for us) can ramp the voltage up.
But I don't think it's all that common for transformer manufacturers to do full rating tests. Those that I've witnessed have had the load losses measured with the output shorted and just enough input voltage to produce full load current. That and the no-load losses are combined.
I can't recall any full power tests that I've witnessed - the cost of the wasted power not to mention the practical difficulty of dissipating all the heat......
Back to back tests are possible with two with the supply providing just the losses for both transformers.