I'd appreciate if you could have a new blogpost with details about the experiment so the community can verify the claims/reproduce it with other DBMS solutions.

Also, while I understand the statement about "lock-free", it seems to suggest the code will be as fast as main memory access, which it most certainly will not (due to cache coherency protocols kicking in while doing cswap, locked-add, etc.) Your comment doesn't say this, but I am noting it here for others.