Looking superficially, I could only find articles claiming the exact opposite – like http://blog.tatham.oddie.com.au/2008/10/13/why-light-text-on... – but they all seem to quote the same and old (1980) research.
Note that not all such studies look at the same thing. I should've been more precise: on paper and with training, people were much faster at quickly reading a page if it was light-on-dark than dark-on-light or dark-on-white.
Looking superficially, I could only find articles claiming the exact opposite – like http://blog.tatham.oddie.com.au/2008/10/13/why-light-text-on... – but they all seem to quote the same and old (1980) research.