The specific questions don't have a hold in real life but the patterns behind them do. Most problems to be solved in computer science, with enough abstraction, have already been solved. I joined a new software development scrum at work and have been helping optimize how long our application takes (it can be thought of as an encoding engine) we've seen huge leaps in improvement with very simple solutions, 75% of the time the way we're seeing that the best way to solve a problem is via a hash table. That's also an incredibly common core solution to most interview problems.
The patterns _can_ have a hold. It's always fun to optimize slow engines, but it always feel more to me like rote/busy work than actually building the system from scratch and engaging in product engineering. But, I guess that's kinda your point -- very simple solutions that revolve around simple, common core structures can lead to huge leaps in improvement. But this kind of real world optimization has medium breadth and depth. It's got a lot more breadth than many algo interviews I've gotten, and yet less depth. I've found that's been the real structural weakness in weak algo interviews I've had -- they just overindex on deep knowledge of algo optimization and underindex on communication, reading and design ability. I'm drawing from a pretty small and outdated sample, though; the last time I had the patience to interview with a BigCo was more than five years ago.
