A recent of example I had that made me dislike systemd was a prototype RHEL7 system here that has ZFS-on-Linux support installed on it.
When you boot it up, there's around a 50/50 chance whether the ZFS filesystems will be mounted after boot. This is an inconsistency that, as a long time sysadmin, REALLY BOTHERS ME.
Yes, I realize the root cause. ZFS has some dependency that is not starting before it. The dependency has to be declared in the appropriate service. However, with systemd we introduce the concept of "just because it came up correctly on this boot doesn't mean it will on the next one."
And that is supremely frustrating. What if it weren't 50/50? What if the likelyhood it didn't come up was 1/100? Suddenly a routine reboot becomes a debugging mission, and I reboot again and it works. "Eh, must have been a transient problem." No it wasn't.
With classic init you were fairly sure that the system's state was the same on every boot. Now it's a gamble. Good luck with that! This is why we're sticking with RHEL6 for the moment on production systems.