Your device is measuring to much harmonics in the second transformer. Since it has more than three times bigger THD (odd harmonics are the critical ones) than the first one, one variant is that the second transformer is supplying non-linear load with greater influx of harmonics than the first transformer. If this is the case, I would be also worried about UPS, since he is also liable to harmonics influence. This might be the case since UPS obviously does not have the characteristics of some modern UPS devices that have clear sinusoidal wave at the input no matter the load properties (kVA=kW, less than 3% distortion at the output). However, if load is not the case, than it is possible that UPS is malfunctioning and therefore does not draw sinusoidal voltage from the network.
What is the transformer coupling? If it has one winding in delta, then the triplen harmonics are captured in it and do not propagate ahead of transformer. That might be the reason for overheating. High I4 suggests that there is also high content of triplen harmonics in the neutral wire.
Be aware that transformer manufacturers use K factor to calculate transformers in correspondence to expected power of non-linear loads. K-factor is a weighting of the harmonic load currents according to their effects on transformer heating, as derived from ANSI/IEEE C57.110. A K-factor of 1.0 indicates a linear load (no harmonics). The higher the K-factor, the greater the harmonic heating effects. When a non-linear load is supplied from a transformer, it is sometimes necessary to derate the transformer capacity to avoid overheating and subsequent insulation failure. The reason for this is that the increased eddy currents caused by the harmonics increase transformer losses and thus generate additional heat. Also, the RMS load current could be much higher than the kVA rating of the load would indicate. Hence, a transformer rated for the expected load will have insufficient capacity.
Generally, only substation transformer manufacturers specify K-factor load de-rating for their products. So, for K-factors higher than 1, the maximum transformer load is de-rated. This might mean that your transformer in not adequate according to new circumstances.
If you identify that the load is the problem, then you should either install filters or try to redistribute loads to burden the transformers equally. You can leave it as it is (if there is no harmonic propagation into the network and you are not paying penalties) and be prepared to change transformer sooner than it is planned, but this time with proper ratings. Also, all the infrastructure down from the transformer will also be overloaded due to harmonics.
