Comment Ho ho ho (Score 3, Informative) 314
FWIW, I'm a PhD student at a reasonably large institution in the US.
Very little of this stuff sees the light of day. The vast majority of software is written simply as a proof of concept for some particular method/system/algorithm in order to get published. Good conferences/journals will typically want not only a well thought out idea, but an idea that you can and have implemented it to some extent, and that it works. That having been said, most of what gets produced is complete and total garbage -- typically just enough code to be able to prove that something runs correctly and in a given amount of time.
Personally, I have written a bunch of junk code during my time here. I'd like to think I know more or less how to write good code after all these years, but writing good, well documented, well tested code takes time we don't have -- writing code is simply a means to an end (publication) -- and so most of the code I write is hasty and ugly. This even applies to code that people say is for "wide distribution".
Before you go hounding on academia however, I'd warn that writing "good code" isn't really the point of what we're doing -- the point is to produce a reasonable method of solving some particular problem or type of problem. Going into bioinformatics for example, there are a whole bunch of problems that involve performing more efficient analysis of certain types of graphs. If a researcher discovers something along these lines, he/she will likely write some junk code to prove that the bare algorithm works, perform some analysis of it, publish it and move on. This may or may not end up actually being a useful improvement -- if it is however, then some implementer whose actual job it is to code whatever medical software might be using this algorithm then has a basic blueprint of how to proceed.
As for some examples of software from academia that have made it out, let me think...
Coverity - static code analysis tool, started at Stanford then moved into being a startup and is now quite successful
PostgreSQL - Originally from Berkeley
Bro (Intrusion Detection System) -- written by a researcher from Berkeley/ICSI -- is still somewhat "in academia", but I have heard of several production deployments
That's all I feel like coming up with right now, but I think the general pattern here is that if/when some piece of software produced in academia is seen to have value in its own right (e.g., away from the original research/publication that spawned it), it typically gets spun off in a start-up or a more concerted effort is given to its development, at which point one can actually spend the time to write good code.