Saving documents is very slow

[bschreck]

Every new document save requires a deepcopy of the entire dataset, since every document contains the _data field. There must be a way around this. Why is deepcopy necessary? Can a reference to the dataset be stored instead?