Since each index on a collection results in an additional insert at the storage layer, we should reduce our maximum batch size when a collection has more indexes. I would propose we take the value of the kInternalInsertMaxBatchSizeDefault parameter double it (because it was tuned assuming one index, _id), and divide by the number of indexes (when we start) plus 1, and perhaps also have a lower limit of kInternalInsertMaxBatchSizeDefault/10.