My Random Write was previously 163K IOPS, I have been tracking this since I upgraded to then-prerelease Win11, and across several HW upgrades (motherboard, CPU, SSD and DRAM). To be more precise I had a record of almost 800K IOPS on Win10, so 575K is not technically a full fix of the regression. However, this is a synthetic microbenchmark. And this large change after this particular fix shows that this benchmark does precisely what the KB article mentions: high contention of concurrent writes to a single file. This is not a very realistic I/O pattern even for (well-optimized) database servers, one should use few threads but many I/O queues. Unfortunately I don't have a history with better benchmarks.
Anyway at this point, considering the large changes that Windows 11's I/O subsystem had to enable multiple features from virtualization to DirectStorage, I believe this is as good as we'll ever get for this kind of microbenchmark of traditional I/O APIs.