For the reference quality, resolution and TC are in principle separate quality criteria. If the temperature is rather stable (e.g. in a lab) 20 ppm/K may still be still acceptable for a 4.5 digit meter, though not good. For outdoor field use I would consider in a problem / weak point.
Comparing ADC's is not that easy, as it is not only the resolution / noise that matters. There are other properties that can be more important, like gain drift, linearity, input aperture, offset drift,... .
The ICL7135 is a typical display oriented one: so 1 ADC step usually corresponds to 1 display step and adjustment / calibration is usually analog with trimmers, possibly in combination with very accurate dividers instead of individual adjustment.
Many modern DMMs use an ADC with higher resolution than the display resolution and than use numerical adjustment by a multiplication in software. SD-ADC chips usually integrate essentially all the time, while the dual slope ones usually only measure during about 1/3 the time - this can give the SD ones an slight advantage when it comes to noise.
The 4.5 digit ICL7135 and similar is a bit more then 15 bits of resolution (+- 19999 counts). With numerical adjustment it should be more than 16 Bits of resolution, more like 18 Bits at least, though some of this might be from oversampling.
Modern low end DMMs are usually build around a special DMM chip set that includes also auto ranging and the input amplifier/buffer and AC functions. Today chances are high to find an SD ADC with numerical adjustment.