Nope. 100% demonstrably untrue.
There’s a few contributors to this thread, not just yourself, that would benefit greatly from a digital audio primer. It’s not a snark, just an observation that being better educated about a subject will help you avoid retelling some of the commonly propagated myths.
This 20 minute video is a good start.
https://xiph.org/video/vid2.shtml
Well, I have found an even better way to through your money:
HP cables, 2 meter, 49k : https://www.thecableco.com/sukhavati-speaker-cable-pair.html
Don’t worry, they also supply alimentation cable at $22,500/m
Also special distinction for https://www.lessloss.com/ which offers a kind of usb firewall which allows usb to be of as good quality as s/pdif or AES (that means nothing at all and in addition it is clear that AES (digital audio professional) is much better, more reliable and more practical than s/pdif (digital consumer audio).
All that to say that when we are not in rational discussion and more in emotional discussions + cognitive bias, there is no limits.
But again, I have nothing against that if that makes people more happy or if they feel that they have a better sound.
Just for fun, I don’t want to hurt any strong audiophile. And if so please apologize.
Torpi
In terms of what we should strive for, there’s an interesting analogy between digital resolution and tape recorder AC bias frequencies, as far as distortion goes anyway.
AC bias is commonly used in tape recorders to oscillate the magnetic domains in the tape coating to make them more susceptible to being aligned correctly (overcoming hysteresis) and thus storing the analogue waveform faithfully. The analogy is that, similar to sampling, you need a bias frequency at least twice as high as the highest recorded frequency in order to avoid intermodulation.
The earliest wire recorders used 40kHz AC bias but, by the time the Ampex 200A was produced (the father of the modern tape recorder) the frequency had risen to 60kHz. This was considered enough at that time but experiments by following designers showed that 100kHz gave better results. Ampex then raised the stakes to 150KHz, as did Studer, while Revox used 120kHz for the A77 and 150kHz for the superior B77.
Yet another example, if you will, of engineering going further than the theoretical minimum in order to achieve superior results.
Questions:
Are the tape recorders employing higher AC bias superior because of the higher bias or are there other factors involved, e.g. build and parts quality, other design differences?
Why stop at 150kHz, if increasing provides “superior” results?
They didn’t. Up to 400KHz was eventually used for audio frequencies. But the point is, surely, that there was no initially obvious reason for going beyond 60KHz and the engineers could have stuck there - except that they found superior performance by going higher.
Thanks for the quick response and for your detailed knowledge of analog tape recorders. Now the only thing missing from this analogy of increased AC bias current in analog tape recorders to increased bit depth in digital audio files is the superior performance. 
But on a slightly more serious note, after 30 years in the engineering business I always felt that one of jobs of an engineer was to try and find the performance sweet spot. The spot where further increases in cost, materials, build quality, etc. did not yield worthwhile gains in performance but where anything less resulted in decreased performance.
That being said, of course there are times when one is simply pursuing the state of the art and costs be damned. However these are special cases and should not be taken as the norm or even necessary for high quality results.
As far as I can tell after posting the original question the general consensus seems to be that 24bit is very useful during the recording, mixing, editing and mastering stages but may be of questionable value at the consumer (end user) level, other than for digital volume control and DSP applications.
Also no one has shown how going from 16bits to 24bits increases the dynamic range of the original recording.
But higher numbers, bro.
With 16bits you have to do all your mastering with such care that 60dB of dynamic range are within 96dB. With 24bits one has very little limits, which is convenient for the industry. Unfortunately, all concentrate on loudness and even when they have 24bits at disposal they push remasterings to the upper limit and compress to levels of wall.
The good old audiophile numbers game. Higher numbers, or in the case of harmonic distortion -lower numbers, are always better except when they’re not, i.e. tubes.
Best comment thus far - the loudness war makes this entire thread irrelevant!
Maybe I’ll start a new thread - “How many bits are needed for death metal?” 
Indded
And there you might be right. Although some may point you to the fact that the 96 dB S/N also means -96 dB of distortion level.
Quite right. After mastering, the general view of the marketing people is that 40dB is about the right level of dynamic range to offer the customer, even for big, dramatic orchestral works. I quizzed a record executive about this and he said that they get customer complaints along the lines of ‘the quiet bits are too quiet and the loud bits too loud’ if they exceed 45dB dynamic range on a CD!
In the rock and pop world, record companies are facing a ‘who gets to sound loudest on the radio’ syndrome and all the effort is put into compression technology to try and squeeze the dynamic range down to 0.5dB - I kid you not! However what I have found surprising is that even in these releases the actual instrument dynamics seem to be very good. If you analyse these recordings with frequency discrimination you can see that, although the overall envelope is held to under 1dB dynamic range, individual instrumental sections have much wider dynamics. The human ear/brain system is able to unravel these dynamics, thanks to the construction of the cochlea, so we shouldn’t be dismissive of the recording or the mastering.
BTW, it is in the mastering, for commercial release, that the dynamic range is decided. I’ve talked to record producers who have complained that the CD masters are often a poor representation of the recording. I live in hope that streaming and MQA will solve this issue.
In the final analysis it is true that there is little point in having greater than 16-bit depth for music’s dynamic range in reproduction apparatus, so the real purpose of 24-bit file depth is to give the high resolution DACs an easier workflow and a lower digital noise floor, as I’ve hinted at. There are real gains, however, to be had in moving from 44kHz to 96kHz, but that’s another discussion entirely and not part of this thread.
If the original music has x dB dynamic range, 16 and 24 bit should reproduce the same x dB dynamic range. If you drop your thoughts on dynamic range for a moment and look at resolution instead it will be easier. As an example, -30dB in 16 bit is represented by 2048 discrete levels. In 24Bit, -30 dB is 524.288 discrete levels. 2048 may be enough, but I think I prefer 24Bit. In my view it is not about dynamic range, but resolution capacity.
Sorry, but no.
The key concept here is quantization noise. With 16 bits, the quantization noise is at -96 dBfs. With 24 bits, the quantization noise is at -144 dBfs.
@ogs This article by Benchmark Audio is a good read, and concurs with @Jacques_Distler’s reply.
Thanks for the link @thumb5. An interesting read. John Siau certainly knows his stuff and is probably right. He fails to explain it properly though. At 0dBFS 16bit has 65.536 discrete levels. Does noise shaping help to provide 65.536 discrete levels at -90dBFS? An un-dithered 24bit system still has 512 discrete levels at -90dBFS. I need to understand more about how dither works.
Good question, and I don’t understand the maths or engineering well enough to know the answer. My hand-waving way of thinking about is it that if you consider digitising (for the sake of argument) a DC signal with a level somewhere between the bit transitions of the ADC, when you add the right amount and kind of noise you’ll get from the ADC something like a pulse density modulated ouput whose average value will give you an estimate of the true (fractional) input value. It sounds reasonable that this could be generalised to varying inputs within a given bandwidth but how much resolution you could then “recover” would presumably depend on the ratio between the input bandwidth and the sampling frequency (and maybe other things). Whether this is the right way of thinking of it, and if so how it’s actually implemented, I have no idea. (Just idle musing, really.)
Interesting way of looking at it. However digital version of a blank picture will have random numbers if bit depth is high enough.
With regard to dynamic range: With a 16 bit recording and great gear, the dynamic range of the recording may (rarely) be a limiting factor. With 20 bit or higher records the gear (and room) are the limiting factors.
With regard to your analogy: Isn’t also a good idea to have some room in the box for packing material of some kind?
24 bits gives exactly that.