Its base and head results were measured on machines with a different CPU, core count or memory, so the change might not come from the code. Run base and head on the same runner type.
The flamegraph wasn't generated: most of the time was spent in runtime internals, which would leave nothing useful to analyze.