On the other hand, I feel software engineering is a slightly different ball game compared to thinking mathematically. E.g. most applied mathematicians who write scripts try not to go overboard on modularizing the code mainly because it is easier to pattern scan with code that fits into a page (and average mathematical models do). Another useful technique is to use the same variable names that were used while deriving the model in the code. While, I am not necessarily defending the practice, I have from painful experience (writing convex optimization code) found that trying to modularize and introduce "meaningful names" to replace those that you just used while deriving out analytically the model and also making sure that your code doesn't go wonky can definitely be a challenge to get right. So yeah, there might be some retraining needed to do at times, depending on the intransigence of the individual involved this might be impossible or fairly easy.
Concerning the other point, it's also true but it's something else entirely. I never meant a mathematician will outdo someone with software engineering education, obviously he will at least have to read about good software engineering practices. I can also confirm your experience, from firsthand account even, as someone who began coding to do basic applied math. Thinking mathematically generally leads to (at best) efficient but unreadable, one-time-use-only code.
Also, beware - correlation fallacy. In my experience it doesn't have anything to do with the education itself, I believe you can easily generalize and say the average person taking up mathematics has a better analytical mind than the average CS major who's in it for the hype.