I guess whether it's overvalued or undervalued depends on whom we're talking about. Your point is well taken, but I think the real problem is that inexperienced programmers want the best of both worlds. In other words, they design systems that require consistency to function correctly because it's easier, but then they do not actually guarantee said consistency, because hey, schemaless is so much faster than boring old SQL.
The way I like to think about it (and mind you, I'm an early-stage kind of guy, so I rarely work on huge systems) is that consistency guarantees give you a lot of leverage, and while it's almost unattainable in the real world, in the computing world we have extremely powerful CPUs and networks that allow communication between any two points on the globe in less than a second. There is frankly a huge number of applications that can be built on an ACID database and never have to worry about the offline or distributed use cases that would require building a proper distributed system.
The way I like to think about it (and mind you, I'm an early-stage kind of guy, so I rarely work on huge systems) is that consistency guarantees give you a lot of leverage, and while it's almost unattainable in the real world, in the computing world we have extremely powerful CPUs and networks that allow communication between any two points on the globe in less than a second. There is frankly a huge number of applications that can be built on an ACID database and never have to worry about the offline or distributed use cases that would require building a proper distributed system.