How will the compiler and specializer jointly handle this case?
A. Quite well indeed.
(To this reader, it sounds like there’s already been lots of sim results and tests that have exercised the existing compiler, specializer and simulator on a whole bunch of loop-heavy code. So there’s a body of NYF/NY-disclosable evidence that the tools and software pipelining of looks works well and will continue to do so. No hand waving involved.)
Well, I could have said “quite well”, but don’t like to brag 
However, I’m sorry to say that we really don’t have the kind of exercise-based evidence you suggest. The prototype compiler never did dataflow analysis (which is necessary for pipelining), and the new one isn’t to that point yet either. So the architecture, and the discussion, is from first principles and not from experimentation.
It is certainly possible that subsequent experimentation will expose flaws in our first-principles analysis, and if so we’ll go back and rethink the analysis. One of our reasons for taking the Mill public as fast as we can get the filings in is so we can get a large number of skeptical eyeballs onto the design; eyeballs that may not be as comprehensive as a test suite but are likely to be already familiar with troublesome corner cases.
We thank you, and your legion of friends who ask us hard questions.