Conflicting results of perf test T4801 in the same MR

The perf test T4801 caused a stat increase and a decrease in the same CI run of the same MR (!2141 (closed)):

Increase https://gitlab.haskell.org/osa1/ghc/-/jobs/202290
Decrease https://gitlab.haskell.org/osa1/ghc/-/jobs/202291

The MR is a very simple refactoring in package db stuff and this is a single-module program, so there shouldn't be any increase or decrease. I suspect this is either an existing non-determinism or an issue with the infra.

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information