Thanks for the response.
1. This would be forging new territory for me too, which could be both a good and a bad thing.
2. This would require a message passing interface, which is not trivial to establish (i.e., they may not be able to test). This is where my skills are most refined and I would be able to defend my answer, plus, it links back to some of my work as an undergraduate (B.S. applied math, did research in numerical methods for problems like this) and shows how my graduate work has matured these core concepts, incorporating more theoretical and challenging CS. I don't know how much you know about solving partial differential equations, but a domain is discretized in time and spatial dimensions and a numerical schema called a stencil is applied as a discrete analogy to the continuous solution. This is straightforward if you are solving the problem serially, but is nontrivial in multiprocessing because depending on the stencil used a communication criterion needs to be established between the processors so that the information is at hand along the processor-domain boundaries.
3. This is probably the most interesting point you've made and I really appreciate hearing it; you're communicating a perspective that I hadn't really thought about, and it makes a lot of sense, especially given my mathematics background.
I think I may be best to tackle 2... Let me know your thoughts.
P.S. Another area that I am trying to sell myself is as an algorithm developer / analyst. These skills can be fleshed out in multiple areas, so I don't think it helps me lean one way or the other, but it is worth noting. Ideally I find a position where I can be part of a development team with a sizable research component.
-- Myles