I just wanted to ask the same question, after doing some tests of my own, here are my findings.
I am not using SSDs but rather standard HDs.
Using the disk managagement mirroring seems to be a little faster. I used queue depth of 8 and came up with values for blocksize 512B, 8KB, 64KB and 4M in Atto Benchmarking Tool:
10M/15M, 140M/155M, 202M/200M, 200M/201M plain disk, only 1 partition
9M/20M, 140M/200M, 200M/200M, 200M/395M plain mirror in disk management, only 1 partition
7M/12M, 133M/151M, 200M/204M, 201M/377M storage pool storage space on first partition, others are a little slower
But.. to me that is neglible, since real performance depends on random I/O (where performance is more similar) and also if you access/share the drives through 1GBit/s network, in which case there is a limit at a little over 100MB/s anyway.
Most important for mirroring is the (rare) error case. So I also googled how to restore a broken mirror. There seems to be a large difference - for whatever reason.
Recovery seems very complicated for storage pool/storage spaces mirror:
For disk management mirror it seems a lot easier
So: disk management way was slightly faster (for me, on Windows Server 2016), plus recovery looks a lot easier. I did not use any cache-size tweaking or anything similar.
I also did not compare CPU usage, since I assume that high CPU and high I/O won't go together very often.
Originally, I also wanted to compare Starwind Virtual SAN and mirror across servers (creating a cluster) but I guess local mirror with backups will be good enough and the performance comparison with that is useless due to network bottleneck.