I messed around with raiding nvme drives when they first came out, about the same time Ryzen cpu's came out (I think). 960 evo's. and then again with 970 pro's. Hell I even did it with the very first gen 1 sata ssd's, same result. Multiple different stripe and block sizes. You can increase q8t1 (best case scenerio) mb/s, and to a much lesser degree, the next couple down. But the one that's the most important unless you spend a lot of time transfering incredibly large files from 1 very fast drive to another very fast drive, is Q1T1, and doesn't change a bit or sometimes goes lower. In every case, you increase latency. There are loads of comparison / benchmarks showing this. If you want speed where it matters the most there's really only 1 thing you can do besides spending thousands on a true raid card with cache on it, and that's to get, even a little bit, more ram and spend the 30 bucks for primo cache or some other similar program. It's real easy to set up. If you use it to increase write speed and use the delay function for even more speed, there is a chance you could corrupt a file if the system is shut down before it completes this process. Personally I write very little, so only use it for read speeds, and have no problems been real happy with it. Hell even if you dont get more ram and just dedicate a small amount like 1gb, that's still a helluva cache to have.
Edit: as far as drive failure, it's incredibly rare either way. That's not the reason I wouldn't do it. I don't do it cause it doesnt do anything for me except if I wanted to show off, half of the results from a benchmark. I also don't do it because it'll be a ton easier to move that drive(s) to a new system and be able to read it right away.