measuring balance error with NTI minilyzer ML1

One customer of mine got pretty strange looking figures when measuring balanced line error from certain high end DAC with his new toy. He mailed me the results and they show 10% error up to about 1kHz pin 3 having the higher level, then the error slowly turns to favoring pin 2 at ca 6 kHz and goes up to 40% (!) at 20 kHz.

I don't mess with semiconductors a lot, so I need some help. Where does this result come from? Bad output topo or wrong kind of termination? I can not believe that the ML1 gives wrong results, because transformer coupled, totally floating outputs from my tube gear showed nice balance. (below 1,5% from 20 to 20kHz) (these were fed from the above mentioned DAC)

I suspect that the DAC has some kind of floating cross-coupled output or something I don't even know the name of, which marketing department can call floating or transformer like. 

The manual of the NTI doesn't really help. The input impedance is around 50k.
So the question would be: does it make any sense to measure the parameter or can these "balanced" IC-beasts really be that bad? Uh oh.

I hope my description is readable  :-\

Some servo style outputs need differential load to work properly. Try connecting 10 kohm resistor between the output pins.

Thanks,

I will try that when I visit the studio next time. I still keep wondering about the poor behavior, one could imagine that the NTI-1 would kind of represent some sort of real world condition. But perhaps it does not.  I suspect that it has 100 k resistances from pin 1 to 2 and 3.

-Jonte