One approach that might help would be to load the models via an XML file, essentially using the shattered loading approach. You can find some info on this workflow here.
There are two variants of that
- Generate a master model server-side. That should have the best performance but requires you to generate this “master model” first on the server (via converter)
- Generate the shattered XML client side and then use the loadSubtreeFromXML function for the loading. The below post has more information on how to generate this XML data (it is written for scs loading but also applies to the SC streaming case):
Model Aggregation via XML
I hope this helps