For Document, BsonDocument, and DBObject Codec.decode implementations, there is a code path for looking up the Codec for a value that requires two lookups in a Map; first by one in BsonTypeClassMap, then followed by one in CodecCache. This could easily be turned into a single lookup in an array that has an entry for each BsonType and a value which is the Codec to use for that BsonType.

Performance benchmarks show that the performance gains would be significant for simple documents.