Do we really need anything greater than 24/96? Opinions?

Going purely based on research evidence, the evidence suggests that Redbook is not sufficient, but 24/96 is: http://www.aes.org/e-lib/browse.cfm?elib=18296
The paper is not perfect (nor the studies it references) and even its justification for 24/96 is quite weak, but with rapidly shrinking storage/bandwidth costs, there is not a lot of reason not to standardize on 24/96.

Somewhere I have a link that showed slightly better timing discrimination in some subjects, with a bandwidth just slightly over 20KHz, but virtually no benefit to going much higher than this. This would also suggest Redbook may not be perfect for everyone, but 24/96 would cover everyone.

You can always take away information at the playback stage if you are worried about distortion at >20KHz.

Indeed there are numerous problems with this AES paper and one of them is raised in the discussion: how can one train for unknown causes? The author’s answer is not convincing to me.
https://secure.aes.org/forum/pubs/journal/?ID=591
Metaresearch can also be burdened with a publication bias; the positive verifications of a hypothesis are more likely to be published than the negative ones. The author writes "The effect is perhaps small and difficult to detect". In the statistical analysis the bias can easily outweigh such observations. Finally, the microscopical population that can hear slightly above 20khz (most adults cannot above 16kHz) is easily served by the Nyquist limit of 22,05 kHz. Anything higher can exacerbate the ultrasonics intermodulation. If reason ruled, all cars should have a factory speed limit of 70mph. But this would impinge on personal liberties, right? Fortunately, in audiophile audio there is no collateral damage, other than that to the wallet and to the ears of an odd dog.

There has never, to my knowledge, been any positive indication that any mechanism in the human auditory system can detect past 22KHz, nor has there been any mechanism shown in the human auditory system that can cause modulation of ultrasonics to audible frequencies.  While this can be shown to occur environmentally, as we are talking a playback system, the playback, if the goal is accuracy, should never add something at playback that was not there at recording.

The question, is, should we avoid the potential for euphonics, for the sake of technical accuracy?  Technical accuracy shuns ultrasonic modulation to audible frequency. That doesn't mean that some (or even a lot) of people won't like it, even if it is near impossible to control. Even allowing that as a creative control is warranted. I think on balance 24/96 makes sense, even if not technically warranted by current research.

So back to the ops question, Yes, 24/96, properly implemented appears to be all we need, and anything higher carries no benefit.

That sounds so authoritative you almost made me think it’s true. 🤗