Like TFA says, you need to look at it like a research OS. One that has critical flaws that are designed in, but that design has a purpose: to function in a different way. Doing so can expose new lines of thinking and novel approaches to "solved" problems. No, it won't function in a real world of networked computing, but it's not supposed to fit into that idiom. It still has some very interesting ideas. Spending time solving every problem again isn't the goal. There was a time when virtual memory didn't exist, for example, and computing still worked well enough to run businesses, banks, telephone networks, and governments. We don't need a research OS to show us that virtual address spaces are useful. We know that already. So, ignore it as not relevant, and do something that is a novel approach.
Looking at the ideas in TempleOS like they will replace Linux or Windows is silly, but they might give us ideas for new types of computing. The idea that everything would be better as a Linux device is, quite honestly, poisonous to the development real progress in the field of computing as a whole.
Maybe TempleOS is like non-Euclidean geometry. Sometimes need drives the development of new math -- Newton's development of Calculus -- and sometimes the math is developed and sits idle, doing nothing for nearly a hundred years before changing the world -- like Boolean algebra. A computer system is just a very complicated set of mathematical rules. Changing the rules of math and seeing what happens has been one of the major forces of change, as different systems are often best expressed in different forms of mathematics.
Does TempleOS make it easier to understand how computers operate? Does it make it easier to learn what a program actually is? Is it just an example of being closer to the bare metal, like you were flipping bit switches on an old Altair 8800? What if, for example, a system like this makes it very easy to model artificial intelligence? It can basically reprogram itself, after all, as everything is JIT and source code is readily available for literally everything at all times. That seems incredibly powerful. Is it possible to write a self-refining program in HolyC?
Just because I don't understand what something could be used to do doesn't mean it's useless. It might just mean I don't have a very good understanding yet. The questions to ask are, "What kind of system benefits from the design of TempleOS? What kind of system benefits from raw, unimpeded access from the user or input to the hardware?"