So I re-compiled the Python GDAL packages to bump the version to 1.11.2, compile in the HDF4 tweak that we required to open more than 32 HDF files simultaneously, and make sure that GDAL works with MODIS data. There's still a glitch, which means that I have compiled the library against GEOS v3.3.3 rather than v3.4.2, but since this isn't a major issue for me at the moment, I won't bother with it at the moment.

Here's how to get things going using the Miniconda Python installation (you can also install all the other packages in Continuum's repositories, but these are the ones that we mostly use):

wget http://repo.continuum.io/miniconda/Miniconda-latest-Linux-x86_64.sh
bash ./Miniconda-latest-Linux-x86_64.sh
export PATH=/opt/anaconda/bin:$PATH
conda install numpy scipy matplotlib ipython ipython-notebook conda-build
conda install scikit-learn pandas spyder numba numexpr cython binstar
conda install -c https://conda.binstar.org/jgomezdans proj4
conda install -c https://conda.binstar.org/jgomezdans hdf4
conda install -c https://conda.binstar.org/jgomezdans gdal
conda install geos

As you can see, the last line install geos again (v3.3.3), otherwise, GDAL complains. This is a minor niggle, and I can open both netcdf and HDF files, as well as GeoTIFFs and so on.