I came across the conclusion here since a change sometimes spans several repos, per-repo history optimizes the wrong target.