I don't agree fully. If some idea looks really good but implementations tend to be very problematic then the idea is likely presented incompletely or inaccurately, because it carries some hidden/non-apparent risk.
Some good-looking ideas almost always result in beneficial implementations, some good-looking ideas almost always result in bad implementations.
If all implementations of a "good" idea are bad then that's a strong indication that the "good" idea might have some significant flaws.
If the "good" idea has some bad implementations as well as some good implementations (like the swedish network example?) then perhaps you shouldn't dismiss the "good" idea so quickly
Sure, let's get to concrete things. What is a separate physical network worth, availability wise? Kind of hard to answer. It depends on the threat model. Even geography.
Some good-looking ideas almost always result in beneficial implementations, some good-looking ideas almost always result in bad implementations.