serpy | ridiculously fast object serialization

I'm developing a GeoDjango app which use the provided WorldBorder model in the tutorial. I also created my own Region model which is tied to WorldBorder. So a WorldBorder/Country can have multiple Regions which has borders (MultiPolygon field) in it too.

I made the API for it using DRF but it's so slow, it takes 16 seconds to load all WorldBorder and Regions in GeoJSON format. The returned JSON size is 10MB though. Is that reasonable?

I even change the serializer to serpy which is way much faster than the DRF GIS serializer but only offers 10% performance improvement.

Turns out after profiling, most of the time is spent in the GIS functions to convert data type in the database to list of coordinates instead of WKT. If I use WKT, the serialization is much faster (1.7s compared to 11.7s, the WKT is only for WorldBorder MultiPolygon, everything else is still in GeoJson)

I also tried to compress the MultiPolygon using ST_SimplifyVW with low tolerance (0.005) to preserve the accuracies, which brings down the JSON size to 1.7 MB. This makes the total load to 3.5s. Of course I can still find which is the best tolerance to balance accuracy and speed.

Below is the profiling data (the sudden increase of queries in the simplified MultiPolygon is due to bad usage of Django QS API to get use of ST_SimplifyVW)

EDIT: I fixed the DB query so the query calls stays the same at 75 queries and as expected, it does not increase the performance significantly.

EDIT: I continued to improve my DB queries. I reduced it to just 8 queries now. As expected, it does not improve that much performance.

Below is profiling for the function calls. I highlight the part which took most of the time. This one is using vanilla DRF GIS implementation.

Below is when I use WKT for one of the MultiPolygon field without ST_SimplifyVW.

Here's the models as requested by @Udi

I have a CSV on the hadoop file system hdfs that I want to convert into multiple serialized java objects using this framework:

Using python 3.5.2 and django 1.11 multitable inheritance like this:

Given two different models, with the same parent base class. Is there any way, using either Django Rest Framework Serializers or serpy, to serialize a chained list containing instances of both the child models?

Given some example models: