Does this suggest that there is merit at looking at the SINAD at multiple frequencies in DAC testing, in general?
Not for this reason. Transformers have a terrible time at low frequencies. They have very different distortion mechanisms to pretty much anything else in audio, and looking at low frequencies is going after where we know they are weak, and where we can separate the good from the bad. Unlike other audio electronic components the integral of the signal matters, as this gives us the level of magnetisation of the core. So as the frequency drops the absolute magnetisation of the core rises and rises. Pretty quickly we get into the highly non-linear part of the core's BH curve, and see not only significant non-linearity, but also hysteresis - so path dependant non-linearity. You can get into arguments about things like thermal lag in power transistors, or retained charge in capacitors, but for the most part there are no other components in audio systems that display significant path dependencies.*
It would have been fun (in a grim way) to see an IMD plot for these transformers. The grass would have been very tall.
Transformers are very linear at low levels, so we don't see the typical higher percentage distortion at low levels and dropping as level increases. In this way they are backwards from typical components. Noise is of course constant, and they can pick up interference easily.
*There are speaker drivers that you can drive into different breakup modes, and this does count as a path dependency. You can drive them into a breakup mode at a given level, and it won't come out of it as the level drops, not until well below the level it first broke up. Yuk.
ETA - voice coil heating is the other obvious path dependency in speakers.