Debug evaluation perf issues (for TeX Live)

I am struggling with texlive.combine: move dependencies to attribute tlDeps, resolve them … by xworld21 · Pull Request #218331 · NixOS/nixpkgs · GitHub, which is restructuring the internals of TeX Live. I get a consistent 10% speed improvement on my machine, but a 5% loss on Ofborg when evaluating all of Nixpkgs.

I can’t figure out what to do from here. Is there any way at all to discover which function calls take the most time? I.e. is there any kind of profiling for Nix expressions that can be used for this?

Cc @veprbl

Just a wild guess: could this depend on configuration of BohemGC?