Thanks for the replies - I'll respond to all of them here.
How do you know they are of poor tolerance ? They are probably NPO dielectric with 1% tolerance or better, much like the resistors
The 270pF caps are 5% tolerance NP0 (not specified in the excerpt, I see), so 50x worse than the 0.1% tolerance resistors (which is specified...), but even if the caps were 1% tolerance that still counts as relatively poor in comparison.
***
The tolerance of the capacitors may not be the biggest contributor. Whats the output impedance of the opamp? How balanced are the impedances seen by the two input paths then? How does that compare to the tolerances of the passives?
Sure, other things can contribute to AC CMRR degradation, but in this specific case I am only interested in the relative merits (or lack thereof) of slapping some 5% capacitors across 0.1% gain-setting resistors.
***
Since the AC CMRR is reduced due to impedance mismatch created by the cap and resistor in parallel, ( so any CMRR reduction is frequency dependent) starting at no effect at DC and slowly increasing to it's max at the corner frequency .
Yep, that's my argument.
If those caps are 1% then at a decade below the -3dB those caps would have an equivalent mismatch effect of those .1% tol resistors, then increasing effect peaking at the -3dB and then above fc CMRR is increasing again as the gain rolls off. The designer decided the small reduction in CMRR over that decade is not a problem, I would probably agree.
Okay, one might argue that the reduction in CMRR from capacitance mismatch is not much of a problem (keeping with your assumption of 1% caps, as the argument is rather less credible when 5% caps are used), but what do you gain in the bargain? My argument is nothing. Tailoring the frequency response should happen on the single-ended output of the diff amp, not in the differential input, simply because it is much more difficult to match the reactances of each arm compared to the resistances (if it wasn't then HV differential scope probes with high bandwidth wouldn't cost so much!).
***
How much does it actually matter? Keep in mind the CMRR goes to pot up near the amp's GBW anyway, so you aren't worsening it much with crummy caps.
A fair question - does it actually matter - but see above response to Kevin.D. And given that 270pF + 24.9k results in a pole at 24kHz, it would seem to me that an imbalance in capacitor values would start degrading CMRR well before the unity gain bandwidth of even the lowly 741 or 324.
Where high CMRR is needed, including through cutoff, a more precise filter is called for (and probably an amp with higher GBW), perhaps incorporating common mode chokes (which can have better balance than pairs of capacitors). If you also need high bandwidth and impedance (such that is impractical to achieve with much or any filtering), good luck...
Well, clearly a high input impedance diff amp isn't needed since the signal source (admittedly only partially shown) is itself an amplifier with a differential output, and also not mentioned is that there is significant common mode noise zipping around inside this box at 50kHz, which the chosen op-amp still has an intrinsic CMRR of around 50dB, according to the datasheet. The goal here is to reject noise at the 50kHz switching frequency of the main power converters, rather than, say, try to maintain gain flatness at >10MHz or so.
***
They should not be used for tailor the bandwidth for the reason you identify and the same goes for capacitors shunting the input. They may be required to suppress RFI or to cancel input capacitance but these occur at much higher frequencies. This also applies for high pass filtering except to provide AC coupling.
If the bandwidth of the signal is to be tailored, then it should be done after the output. The not so obvious way, especially for a high pass response, is to apply the filtered output to the instrumentation amplifier's reference input.
Yep, I've always applied filtering after any differential-to-single-ended conversion, whether using either a single op-amp or a true instrumentation amp to perform the latter. Admittedly, I didn't give this much thought until the current situation popped up, but it - and an earlier reply from T3sl4co1l about a current sense amplifier circuit I am fond of using - caused me to rethink (or think about) some long-held assumptions. Thanks for the replies, everyone; I think I'll do a treez and give you all a thanks for participating...