Are the claims a large part of the network traffic for items you are
processing? Some client time might be saved by lazy loading the claim
objects from _content. The claims data is even smaller when using raw
revisions instead of the API JSON.
Is the size of the serialization something that is causing problems?