I decided to record some 60fps footage of Resogun to see how long it takes for the OS to become active to user input after the home button is pressed.
I ran the video through Final Cut Pro to count the frame.
If you take a single frame as 16.67ms, and multiply it by the the number of frames, in theory you'll get the total transition time.
I was able to count 25 frames going to the menu from Resogun and 27 frames from the menu to game. If we take the average here, we'll get:
26*16.67 = 433ms or .433 seconds.
Cerny said you can transfer 2GB in .27 seconds in this slide:
NXGamer
and his theory about OS/RAM is entirely feasible within the time I've calculated and certainly a large chunk of OS data could be loaded in on-the-fly. Sony would actually have ample time to spare and make that transition even faster than it is on PS4, that is based on the assumption that it's 2GB of RAM they're saving and that is the size of OS RAM footprint, excluding the OS kernel and subsystems which would need to stay in active memory.
Basically, all non-essential RAM could be purged from the OS, thus leaving significantly more RAM for devs to use in games.
Let's look at the the best-case scenario for XSX's SSD speed, all else being equal. The SSD there is 56% slower than PS5's so within that .27 seconds, they'd be able to transfer 1.12GB. To transfer 2GB, XSX would need .618 seconds which is far outside of the .433 seconds needed for PS4's OS. XSX is reserving 3.5GB of RAM for its OS. If they try to dump 3GB of OS RAM to storage at speeds of 2.4GB/s, it would take well over a second, which would cause their OS and UI to feel laggy in comparison.
All of this is of course only theoretical not taking into account any latency reduction measures which look to be present on PS5, but unknown on XSX.