I'd like to know on what machine and on what testbench the supposed 300x boost is achieved so that we can independently verify it. My assembler-written fork of postgres is achieving 600x boost on SELECT 1;