It isn't so much that "old is bad" as that the new is more likely to have been designed with modern paradigms in mind. Despite your dismissal, parallelism in particular is important, especially as Linux has taken a role as the embedded OS of choice for smart devices and cheap laptops.
Linux was succeeding quite well in the non-RTOS embedded space and with cheap hardware long before systemd came around. And an embedded device (aka, an appliance) is precisely where you want the MOST deterministic functioning. You don't just randomly through a bunch of parallelized shit in there and hope systemd all figures it out for you.
The fact remains that the previous init harness was perfectly reasonable. People that needed service management, socket launching, and other functions had options in daemontools, inetd/xinetd, runit, and myriad other tools out there, while rc (on BSD) or the chkconfig-controlled symlinks in rc.d gave structured sanity to the set of deterministic instructions a machine needed at boot.
Systemd's writers forcefully shoved all that aside in favor of a one-size-fits-all strategy that people had to accept whether they liked it or not, and once it was in place, they did everything they could to burn the bridges back to other paradigms.