http://tayloroshan.github.io/sparse_glm.html
PySal has a very interesting GSOC project for sparse spatial GLM models with emphasis on Poisson (according to the proposal).
This is a bit similar to what we discussed and I started a few months ago with absorbing a large number of fixed effects or categorical variables. However, I looked only at the linear case.
I expect that Taylor will produce some interesting results that we can bring into statsmodels to improve our sparse support.
I haven't looked at the details, I just saw his blog post.
Josef