Most motherboards have 32Gb/s dedicated to one or two m.2 drives so if two m.2's were raided it may bump up against the limited of the board. Even one m.2 drive will hit 20Gb/s on it's own.
This seems like the right trail to follow, even if Gigatons didn't fully explore it.
Your motherboard would be relevant for its limitations. If you have a LGA-2066 or LGA-2011-3 CPU and motherboard, there should be ample bandwidth for additional NVME M.2 drives, and additional sockets for them. While I still think of my Skylake and Z170 board as "new," it is the boundary of my personal experience.
If you lack motherboard M.2 or U.2 sockets (do I have that right?) -- you would be using a PCIE x4 card for any single NVME. I can't be absolutely sure you could RAID NVMe's in two different sockets. If you want them to run under a single socket, then you need a special PCIE adapter and x8 bandwidth for it. But you also need the motherboard to support PCIE "bifurcation."
With my Z170 board, I have two PCIE sockets that operate at x16 or x8/x8, and I have a third socket configurable for x4. With a graphics card, there's only one socket left. I use it for a second PCIE x4 NVME used as cache for SATA drives. The board doesn't support bifurcation.
But since I can cache NVME drives and SATA drives to 16GB of RAM while caching SATA to NVME, my benchie scores for -- say -- sequential read are in a range between 11,000 and 12,000 MB/s. And my 4K random read scores for SATA devices is somewhere between 600 and 700 MB/s. I am still wondering what it would be like if I replaced the 960 EVO caching NVME with an Intel (Optane) 900P. But my wallet and my happiness tell me not to waste the money.