I've found that s6 and s6-rc are correct to a fault. It works as documented, but you can't take shortcuts like you can with other init systems that don't have a validating compiler compilation step.
"Aggressive" processes that for one reason or another escape well-mannered foreground supervision; for example, often those that daemonize themselves incorrectly. Either they don't generate PID files correctly, or they don't cleanup after old PID files leaving them dangling, or they have useless PID files that point to the incorrect child processes that can't do useful work when they receive signals to reload/shutdown themselves safely.
Handling those daemons by jailing them with Linux-centric cgroups is the one thing systemd allows you to "get away with".
