This one doesn't seem very important to me, or rather I see an altogether different design as more resilient. Instead of trying to achieve a fewer (by economic necessity) number of highly reliable nodes, we can have a larger number of nodes, even if any one of them is relatively less reliable, where each node can take over and cover for any downtime of another node. This would make the network as a whole highly more resilient and robust.
No, unless you have entirely different consensus algorithm in mind. But we have this one, and reliability of block production among consensus witnesses is exactly what it makes it resilient and robust. :-)
Oh, right, thanks for pointing it out.