In our current 2.0 design, a new serialization context must be created every time we descend one level in the hierarchy of the document being serialized or deserialized.

Most of the time, the only thing that is changing is the nominal type. Perhaps it would be better to move the nominal type out of the serialization context and back as a parameter of the Serialize/Deserialize methods, which would allow us to use the same context over and over, thus reducing GC pressure.