T18623 fails consistently with "failed to create OS thread: Cannot allocate memory"
This has happened in a few pipelines recently.
Best example is https://gitlab.haskell.org/ghc/ghc/-/pipelines/64429, where nightly-x86_64-linux-deb10-no_tntc-validate
failed repeatedly on that test.
On that pipeline, the job was retried numerous times because "Cannot allocate memory" is my match string for an out-of-memory spurious failure. The fact that this particular test is failing over and over, across runners, makes me think it is not spurious in this case.