I know of two ways currently being tried to address this. The first is to have scientists break their problems down into chunks that can be performed by volunt

I know of two ways currently being tried to address this. The first is to have scientists break their problems down into chunks that can be performed by volunteers but which aren't completely beyond the ability of the scientist to manage the resultant code. This is doable for much of physics and computational biology, less so where a scientist isn't a programmer themselves. We're taking this approach with http://solvers.io http://solvers.io. The second is to have the scientists mentored by programmers to help them become better at it. This is the approach being taken by http://interdisciplinaryprogramming.com http://interdisciplinaryprogramming.com. In both cases, any particular volunteer dropping out is probably not a massive blow. If a project is going to rely on a particular programmer long-term, they probably need to find the funding to pay them.