Did you try increasing the full-scale voltage one notch on the DVMs to see if you had any crest-factor problems?
Excellent point--I did on the bench meters and had some slight variations either way, except for the 8506A which can handle insane crest factors without any issue. With all the other ones it was difficult to discern the difference between CF and BW variations between ranges. Using autorange, these are the ranges that the meters settled into.
8506A: "3V" range, full-scale of 4.000000V, full rated accuracy with up to 8:1 CF, peaks less than 16V and HF components all less than 10MHz.
8505A: 10V range, CF <7.
8842A: 20v range
34401A: 10V range, 12V full scale
8808A: 20V range.
So you can see those should have no issues with peaks going out of range and the CF is actually less than 2 for this waveform. However, the handhelds showed a little improvement and both read a bit over 3.0V on their 50V ranges. Still some of this is BW (both meters are only specified to 100kHz) and some is CF (specified as <3.0 at FS). To sort that out I tried slowing the waveform down to 12kHz and then I got:
Fluke 289 (5V range) 3.16xx V (50V range) 3.206 V
Fluke 189 (5V range) 3.17xx V (50V range) 3.211 V
Those readings are within the specified tolerances for both of those ranges once you do the math. So the issue with the handhelds and the 120kHz signal seems to be mostly BW or perhaps some combination.