Suspicious inactivity kills

Simple tests like https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/f07867e_vs_c3687a8/report.html, using a stable Julia, include a lot of inactivity kills during or right after precompilation. I wonder if we're doing something wrong detecting these, e.g., not correctly incorporating the CPU usage of child processes.