I'm not sure why the op-amps like these faded away instead of getting better. The main input amplifier on this unit is a discrete high input impedance op-amp with separately adjustable balance and bias. The mV preamp uses the ADO-267B which only has adjustable balance, the bias ends up being whatever it is. Decades later Fluke was still building their top-of-the-line 8505A/8506A with a discrete op-amp front end with similar adjustments and I presume similar design. If that ADO-2xB had been expanded and developed, it probably could have take the place of that whole discrete op-amp, since the performance appears to be quite good as it is. Instead it appears to have gone away. Perhaps it was very expensive?
I think they were just too expensive.
Integrated JFET and CMOS parts never achieved the performance of discrete ones but you can still get that performance and better today doing what you originally suggested; select a dual JFET to make a composite amplifier. The integrated route using a part like the OPA627 or one of the other parts I mentioned is just much simpler and cheaper if its performance is acceptable.
It may not actually be necessary to match or exceed the performance of the ADO-267B in all aspects but this would take some reverse engineering to figure out what is important and the same list of viable replacements would come up anyway.
I have a number of old digital multimeters with discrete front ends which perform amazingly well even after 40 years, and improving on their performance with an integrated JFET or CMOS part would be difficult, except where a super-beta precision bipolar part could be used like the LT1008 or LT1012; they are like magic.
I try to resist the urge to improve most of the old-tech things I work on, not because I'm a stickler for originality, but rather the realization that most of these things were made very carefully by very smart people. Also, a lot of these things were quite advanced and expensive in their time and while it looks like you can outdo them with modern equivalents for pennies on the dollar, that's often not really true.
In the case of the DM501A that I am restoring, the design was flawed and honestly I think the designers at Tektronix should have recognized the flaw in the ADC chipset they were using either during design or during testing. Eventually it was identified and fixed, although Siliconix never owned up to it that I found, but the solution was less than ideal as I outlined above because it depended on the CMRR of a JFET operational amplifier.
When I do reverse engineering, I often find provable mistakes.
What is interesting to me about this old (1967 or so) DMM is how well it actually performs, or hopefully will once I get it straightened out. It is only 12,000 counts, but it has high-impedance inputs to 12V, <10pA bias current (the max spec is <50pA, <100pA on the mV range) and the accuracy is +/- 1 count offset, +/- 1 count reading at full range and +/- 1 count per 5C tempco. So essentially a max error of 2 counts over the entire range.
"Only* 12,000 counts - 12,000 counts with only 2 counts of error is a credible achievement and most modern comparable multimeters are not that good.