I've never understand that argument either. I suppose it might make a difference if you are constantly (shutting down and) spinning up new instances all day long. Most of my hosts "boot up", at most, two or three times a year, though. An extra 30 seconds is not something that causes me pain.
Even in a cloud environment where you might be scaling dynamically based on demand, it's fairly trivial to set your scaling policies with just slightly lower tolerance to account for a couple dozen seconds startup time. If you're not scaling until you have an immediate need for that capacity, you're in trouble and your users are going to be encountering reduced performance or service outages anyway.
It's important because systemd boasts faster boot times and to show that systemd isn't really much faster than existing solutions. Boot times are important if you start on demand. See MirageOS's network initiated quick boot for one extreme.