I'm a PhD student in mechanical engineering without an extensive background in computer science. We write code as part of our research, but it's usually high-level (e.g., Matlab) and often rather ad-hoc.
It sounds like code reviews amongst the academics here would be valuable for (a) learning techniques from other people, and (b) spotting inadequacies in your code. (Sometimes I think it's scary that there are papers being published where the theory is instantiated in code that no-one but the author ever looks at!) Should academics have mandatory code reviews with their peers? (Has anyone seen such a thing before?)
More importantly, how long does it take to do an effective code review in this context? It seems hard to justify the extra time when that particular resource is always very scarce.
Addendum: A couple of people have asked whether "publication" constitutes a sufficient review. Not at all, in my field. The results are the important part, so if you write your algorithm and/or data analysis with incomprehensible code that spits out a graph after hours of processing, that's no different than clean, sharable, fast code.
You might write buggy code that produces results that look about right and never know there's a bug in there, though!
But I think it would be a great idea to have publications distributed electronically along with the code that was used to produce the results in the manuscript. Problem, again, is time: everyone's code is ugly and unmaintainable (to generalise; afterall we're engineers, not programmers), so cleaning it up for publication would take too long. Code reviews might help this situation, a little.