Hi,
the warning for deployment of GeoDjango seems to me a little over the top. Wouldn’t it be enough, to put locks around the code that deals with geo data? Increasing the process number is not cheap…
Frankly, I have several (Geo)Django instances running in multithread context (not so much traffic though) and I never encountered issues related to GDAL. My feeling is that issues may arise when using the threading module with cross-manipulation of geometries in shared-memory context, not really in different threads handling different requests. But that’s only a guess…