This is one of the areas where I believe source release is certainly noxious. We are talking complex science code. The (public) algorithm are already hard to prove, hard for peers to review, it is very easy to spoil a good algorithm with a nasty bug in the implementation (especially when you are trying to make it superfast by cutting the corners).

Having multiple teams recode the algorithm at the same time and eventually confront the results on similar data sets is a safeguard against this. Copying the code on one another would be too tempting if it was made public immediately including the build scripts. Of course, it should be made public eventually, because proof reading can uncover yet more bugs. But the "full transparency" is probably flawed.

Note that the same may happen with more "commercial" software like open source OS and middleware. Several competing projects must be kept alive, and they must recode stuff on their own, just to keep the "software gene pool" from shrinking. Having just one libc over the entire planet would make a killer of just one single buffer overflow exploit...

Pierre