Software engineering, on the other hand is too important a subject to be left to the schools. Let them learn Scheme on their own. They'll appreciate it more that way.
Language details are all academic, pointless debates to be had by people who like one over another where the differences are often trivial (C++/Java). If you have a great mind and can understand what languages are doing then either will do you just fine.
I'd rather not teach the next generation of CS under-graduates the same old stuff that I did (and yes that included how the VAX-11/750 - one of the first to do so if I recall correctly - didn't have to have all a programs data or code in memory...)
Teach them things that will take CS forward...
Among the higher-level problems you mentioned, how many of them are interrelated? I'd rather see professors cover material that is useful in 80% of cases, even if they are well-trodden topics - students can specialise in the rest. In my CS course (UK) we studied parallel systems, including MapReduce; distributed databases and NLP can be studied optionally along with several other topics of much higher level than mere page faults. That said, not all CS courses are created equal.
you do not need to understand how your compiler works, or how De Morgans law is responsible for all those NOT gates in your CPU to produce useful and meaningful CS.
"Programmers who don't understand the memory hierarchy write slow code" is just nonsense.
Most programmers today use languages where they aren't even aware of, or able to manipulate in any way how and when memory is managed. And this trend will continue as the hardware their programs run on are increasingly virtualized.
Let's face it, virtual memory is a done deal, its now fundamental until we find computer hardware architectures that don't ever have address spaces that exceed the physically addressable memory. With virtualized kernels, possibly its a better solution to boot operating systems that have no concept of constained memory resources, and let the virtualizer do the paging.
The problem with Java (and C, to a lesser extent) in basic problem solving is that you can almost brute-force-code and get a working solution that will get full credit, but after writing it a more elegant (and efficient) solution isn't always obvious. In functional languages, that "more correct" solution almost always seems to stand out more, at least to me.
If a functional language was taught in beginner CS classes I think it would be easier for students to see how it works. It certainly "clicked" for me once I'd learned a bit of Scheme in my first AI class.