Hi Gary, Jim, All,
I ran a test like that about 7-10 years ago using a pair of Triplett 630-A meters on the HV side (they have a top 6KV AC scale), and a Radio Shack 22-147B on the low voltage side to measure the respose in the "forward diection". I ran 0 to 48 V input using one 630-A then 45V to 95V input using a 630-A on each side of ground. These two plots were very colinear. I then tested it "backwards" using two of the RS 22-147Bs from 140V down to 0 V on the secondary and got a second plot which was fairly linear, but NOT Co-linear with the first two.
The 147Bs were calibrated before and after against a Fluke 77 series II and showed less than 1% deviation over the 0-140 V range.
I suspect that putting low voltages on the secondary may not create the same magnetizing effect as a low voltage on the primary due to the physical layout of coils and core especially with current limiting gaps. But if one is looking for only a "ballpark number", the reverse method should be OK.(~10-15%).