I live in Norway. Most people here drive stick-shifts. I can guarantee you that the vast majority of people driving stick-shifts have no idea what synchro rings are, nor would they be able to point them out if you cracked a gearbox open and told them to point them out. So yes, the simile is flawed.
As for the usefulness of learning C or C++, I do think it is useful as a learning exercise. As would assembly programming be.
I have met developers who considered themselves "experienced" who are unable to estimate memory requirements for simple data structures. I've had people tell me that pointers are "about one byte" and people who insist that the most compact way to store a boolean value is as an int value.
More often than not it is not the language that decides the performance of a system, but the ability of the programmer to come up with sensible designs.
This becomes even more important as parallel and computing becomes more important. Designing and implementing systems that run on a single CPU isn't all that hard given all the tools, libraries and abundance of literature developers have access to at little or no cost. Writing software that spans CPUs and even computers is a different matter.
Writing an application like Twitter isn't hard. Making it scale shouldn't be, but apparently still is. Doing twitter in C++ might offset the problems they experience somewhat, but not meaningfully so.
By today's standards, your surveillance solution would qualify as an embedded solution and a fairly resource constrained one at that. It would make sense to use C or C++ in that situation: you can afford to spent a lot of effort on a fairly minimal set of features.
The same may not apply for other projects. I for one spend most of my time worrying about IO, availability and how to best take advantage of multiple CPUs and cores. In this, C++ isn't exactly giving developers anything for free (concurrency still not even being part of the language).
As for Java on mobile: there are projects that are going that route. I'd wait and see before I postulated whether this is a good idea or not.
I don't really understand how Vista fits into this discussion at all. In my opinion, Vista is the result of developers being paralyzed by the sheer weight of their legacy code and interoperability with legacy code. Completely different issue.
The main difference being that finding the memory leaks in Java, or rather: unintentional object retention, is far simpler than in most C++ environments.