The RK3588 based boards are PCIe3.0 and providing whoever makes them x4 is 'normal'. The OrangePi5+ I am playing with is heading for an old Nvidia Quadro GPU next on a NVMe - PCIE riser card as there is factory ARM64 drivers to play with.
The dual drive thing is for me just a plaything to see if it can be done
Fun stuff. I tried a dual M.2 Riser card in a Dell Optiplex and got 1 drive. Expected.
I then tried the same dual riser card in a brand new B550 Asus Prime board and .... 1 drive. No 4x2 bifurcation support.
I then tried it in the X570 Crosshair Hero 8. Along side the Gtx3080 in x8 mode and it bifurcated it and gave me 2x full Gen4 M.2 drives.
I promptly put those into a "Windows striped volume" and use it for "Steam games" and video editing scratch drive. It's completely unreliable and a single corruption takes out both disks, but it has "replacable class" media on it. The speed on krystal disk mark is 11,700MB.s. 11 Giga bytes per second.
IRL though. Game loading and scratch driving produce a large number of "random access reads", which, surprisingly (?) will drop an M.2 down to a tenth of it's full speed at sequential. I believe that is the DRAM cache on the SSD working at it's best in "read ahead" mode. When you start hitting it with random access for single blocks, I believe it crashes out the cache on the drives and bottle necks them. Put another way. A game which took 1 minute to load of a "consumer 500Mb SSD", still takes 35 seconds to load on a supposed 11Gb/s M.2.
Edit: There is talk about games starting to be optimised for performance SSDs. Previously game devs moved to "cabinet" or monolithical "dat" files to speed up loading due to seek times on HDDs. When SSDs became the norm and the seek time went from microseconds to picoseconds, the became lazy and went back to "catalog" folder structures and millions of small assests.
It seems they are having to go back the other way as gamers are realising an SSD/M.2 hit with constant random reads does not perform well. With modern consoles getting M.2 this time round, the "talk" is in a redesign (again) or assest management to improve load times.