When we build distributed systems, the need to setup a distributed consensus algorithm is appearing in front of us, time and again. Leslie Lamport (of LaTeX & Time-Clocks fame) came up with a novel algorithm during early 90s about to solve this is a very competitive way (Paxos is its name). Sadly, the algorithm remained shunned for a number of years, due to rejection via the very same channel in which it was eventually published many years later. If you realise the immediate practical impact of that algorithm and what an 8 years delay means in the world of CS, and the cost putting all these together, the result is staggering and sobering at the same time.
So, yes, let's now all peer-review this statement: "peer-review systems are imperfect and provide no guarantee for any certain quality result".
Peer review is merely a compromise to increase throughput of papers, which are relatively median and more easily digestible, because this is what keeps the academia salary system in good lubrication. It provides no level of assurance that the most impactful paper gets noticed first, neither that it receives sufficient feedback in order to improve upon original concepts. In sort, human intellect won't be easily replaced via a procedural setup, yet.