There is a development branch of Statsmodels that allows SARIMA (optionally with exogenous regressors as well) via state-space models. It won't be integrated into the core until after Statsmodels 0.6 is released (the release should be this month, then the state-space branch integrated in the upcoming months).
In the meantime, if you would like to do this (and are willing to install / build a development branch of Statsmodels from source) then you can use the code found at: https://github.com/ChadFulton/statsmodels/tree/kalman
To do so, you will need to have Numpy, Scipy, Cython, Pandas and Statsmodels available on your system, download the source code from the Github repository, and then in the downloaded "statsmodels" folder, issue the command
python setup.py install. This will overwrite your default statsmodels (so if you don't want that, you might consider making a virtualenv)
If you do manage to get it installed, here's an example notebook showing how to use the SARIMAX class: