DA> I saw that paper, but it lacks specific examples of how anything 

True enough.  Once you get to moderately complex, however, I think you are going to just have to get the book....or trick someone into mapping it out.

Would you like to volunteer to put something preliminary together 
DA> that uses IMAKE? Then we could evaluate it against other options.
I guess I should have seen that freight train coming before I got on the tracks.... 

I'm pretty busy, but I may be able to find a few hours over the next couple weeks to try out a demonstration. That is, unless someone else comes up with a better strategy in the next couple days. 
Of course, IMAKE isn't going to solve that runtime validation which I think was part of your original query and I think is really key.  To make Boost really useful I want to compile Boost with all optimizations (or whatever combination I choose) with whatever compiler, on whatever OS, and still be reasonably sure that Boost still runs correctly...ideally by just running the test suite.  Perhaps, however, that should be a different discussion thread.

Yes. As far as I can tell, even a Make expert has to "learn a new
DA> make system" for each new project.
I would agree that the developer of a Makefile has to understand a different system, but the user doesn't need to know much.  For example, I usually expect to be able to change to a directory and type make (or in VCC hit the "build" button) and have everything that "needs to" compile just will.  I expect to have a way to clean up all derived files (make clean), etc.