I mentioned elsewhere in this thread that I did an EE/CS dual degree after having programmed for a long time beforehand. I really don't understand why, but some of the EE-only software I've worked with has been absolutely atrocious.
One of my favourite moments in undergrad was in my Digital Communications class. I don't remember exactly what we were doing, but it involved Matlab and some kind of signal decoding. I handed in my assignment and got called in to have a one-on-one with the professor; he was certain that I had cheated and just hard-coded the answers to my assignment instead of actually writing the decoder. Why? Because my code ran instantaneously instead of sitting and grinding for 30+ seconds. What was my secret? I'd seen a couple of triply-nested for loops in the code and decided they could all be accomplished with a matrix multiply instead of doing all of that computation by hand.
In the same class, a friend (also an EE/CS dual) and I came up with the dynamic programming solution to the Viterbi decoding problem as well, and again, got taken to task for having a solution that ran too fast. Sigh...
Be warned that FANUC is very litigious about that sort of thing. I can't go into specifics, but they've sued companies out of existence for less.
In case you haven't seen them, the SysInternals tools can serve as an interesting middle ground before popping the DLLs into an assembler. There's one (whose name escapes me) that is basically strace-for-Windows, and I've used it to successfully solve some pretty gnarly Win32 problems before.