Lets assume for the sake of argument that logind (a new codebase) is about as stable as syslog (anyone of which are several years old, well tested). Layering one on top of the other, you now have doubled the chance of one of them going belly up.

When have you ever seen syslog go belly up?

You are clearly too young to remember. But don't worry journal will get there

That's the entire point. Syslog has years of stability and reliability. You don't just throw that away, especially for such an important role.

Can you assure it is 100% failure proof in all UNIX variants that have one running?