Having used these devices for too many years, I never had issues with large output capacitors (1000-4700uF). If this seems a lot, imagine the devices are connected to and powering a small mixing console, with 470 uF local bypass on each rail on 24 channels, a typical application for them in the 80's. However I would never use an output capacitor of any size without the reverse bypass diode across the device. The diode prevents device failure due to reverse voltage if the input is shorted or the input rails go down quicker than the outputs on power-off. And I would never in any circumstances use a tantalum capacitor anywhere ever across a power rail if I wanted to keep my hair... (wait!) They are the most unreliable components ever made IMHO. 317/337 were convenient back in the day, but they are seriously ancient tech. Surprised they are still used. The 337 is not very good, a negative rail built with a 317 is more robust IME. Oh, and these devices are not LDO, need 3V across them to regulate...
I agree with most of your comments, including working w these guys for... erhm... a lot of decades! ;-)
Thoughts...
Ancient Tech, yep thats true, but hey, they work well, and they're cheap! With that bypass cap on the regulator pin, they are significantly quieter than a basic 78xx fixed reg. And, it looks like Amir proved that the changes didn't result in meaningful differences in overall noise.
I've also seen many circuits with >1000uf on the output without issues. I don't know where the modder is getting the "shouldn't have too low of an ESR on the output" idea, it doesn't seem to be in the datasheet! If you have LDO regs, now they DO have sensitivity to output cap ESR, and may oscillate if its wrong... but its usually because ESR is too high.
Measuring fast spikes, and in the <10mV range on a scope is a tricky thing, without special equip and techniques. Do you have a high speed differential probe? Did you rig up a way to minimize the ground connection to as short (few mm) as possible, or do you have the usual 4" ground lead? All the difference shown could be just how the probe and ground clip laid out on the board differently, the second time it was measured. Really, if you want to try to minimize high freq spikes, you WOULD put small value, low ESR caps across the output rail, and of course around the circuit, as they likely are.
On the diodes, well, yes its safer. But, the output, inside of a larger device like this, shouldn't ever get shorted, vs you are making a bench power supply, where there a likely chance it would. Prove to yourself your circuit won't ever look like a short and you can drop that. And if you see that you always drain output before input, you can drop the other. Id also have to say in most commercial circuits I've seen, the diodes werent implemented. Belt and suspenders, vs save costs. I would imagine Jason looked that over and deemed it safe, and went on with the another design concern... keeping the Magni at the low price point.
Jason could have put in some of the sexy, ultra low noise LDO parts that are available, had >10X less noise.... with parts that maybe cost 5-10X, and need special output caps, also 5-10X... but would you notice, in the final amp performance? (Or even be able to measure it?)
Or... use a venerable set of regulators that are cheap and good enough because of the high PSR of the rest of the circuit, and have it work pretty much as well for $20 less in final price?
I think a lot of the "picking on the designer" thinking loses track of the bigger picture, sometimes. ;-)