As a one-time developer of codes for climate models (these days, I'm more of a user), I would like to draw your readers' attention to a peer-reviewed journal that explicitly tackles many of the issues Nick Barnes raises (Nature 467, 753; 2010).
The journal Geoscientific Model Development (http://www.geoscientific-model-development.net) strongly encourages publication of modelling codes alongside detailed descriptions of models. It was founded because models are seldom subject to the same degree of scrutiny and peer review as the results they generate — even though modelling is central to research into climate science.
As Barnes so rightly states, model codes themselves are rarely or never published traditionally. Model descriptions often need scientific results for publication and so are pared to the minimum when they do appear. Issues of reproducibility, platform dependence, version proliferation and the real nitty-gritty of modelling all need to be addressed in the literature.