With a 3000 MB/s drive, you can for instance, INSTANTLY play the presets in Omnisphere as you toggle from one to the next. When you double click on a Preset in Spitfire BBCSymphonic Orchestra PRO, the entire 400MB sound is ready to play by the time you get your hands from the mouse button to your MIDI keyboard. No glitches, no missing notes. Double Click on a Full Drumset in NI Komplete and the whole drumset including the cymbals is loaded in under one second. If you have your Kontakt instances purged in VEPro, when you select that midi track in your DAW and start playing a melody, or whatever, there aren't the glitches that are there with an SSD drive as it quickly loads the smaples for the notes you are playing. All of those things for me, are actually more fantastic, and more appreciated then any improvement in loading times of sessions etc. It's like having a hardware synth where the sounds play the moment you switch from one preset to another.
I did a few tests here on the M2 macbook air and my Ryzen 9 - interesting actually.
The M2's speed according to Black Magic is around 3000. The Ryzen 9's C drive (Crucial T500) is 7500. One sample drive E is a 4TB NVMe (Crucial P3), and that benches 3500. However, the Black Magic only returns one test result, which I think is CrystalDiskMark's SEQ1M Q8T1. Some wild variations in the other tests:
(my other drives are regular SSDs for samples, and rust drives for projects)
In the real world, my C drive seems very fast, I boot in a fraction of the time I did. Cubase projects are loading at about twice the speed. However like jbuhler above, I'm not sure there's a lot of difference in Kontakt load times or streaming samples. Omnisphere isn't instant as you find, but it is massively quicker - though part of that is that my old rig had the Omnisphere slow GUI bug that plagued some Windows users that they only recently just fixed.
Interestingly, Spectrasonics support said tests 2 and 4 are the most important for their software. Given the colossal difference in test 2, it might be worth be experimenting putting some stuff on my C drive (it's 2TB so quite a bit of room to spare). I think the days of keeping program and sample drives separate are gone, right? That said, performance is now very good, light years above what it was, so why rock the boat.
EDIT - apologies as I seem to have dragged the thread a little off topic, but where there is an overlap is in knowing what tests to run and how to interpret them.