Most servers boot really slowly anyway (memory tests, raid cards etc), so why would anyone care if it takes one or ten seconds to boot Linux when the server boots in 120 seconds. Other init systems are capable of restarting things with dependencies as well, like BSD's rcng for instance. I am using runit to supervise my services since I can't rely on systemd.
My experience with sysvinit is that it definitely takes well over ten seconds to boot Linux and start all services, especially if you've been putting "sleep 60" into certain scripts to make boot ordering reliable.
Again, I'm comparing systemd to pre-systemd Linux distros (including Devuan, which explicitly continues that tradition), not to systems broadly comparable to systemd. If Devuan had a different fancy init system or service manager (which could just be bundling runit configuration for everything the OS ships with) instead of an old-school init system + old-school init scripts, this would be a very different comparison.
Startup order (again, with traditional runlevels and nothing like insserv) isn't enough to express "Once networking is up, start this job" or "Once this service is ready to accept connections, start this other servie".
Or containers! There are definite advantages to having your container's pid 1 be an actual init process (rkt-style) instead of the application being contained (Docker-style), and there are often advantages in that init being a full-featured init instead of a dumb EWONTFIX-style init. (Although I have written and used the EWONTFIX-style init for other purposes, where that led to a simpler architecture.)
