Support split documents on disk
As an application's functionality grows, if kept unchecked it is easy for average document sizes to grow simultaneously.
Large documents can have a significant impact on the performance of MongoDB. However, often the application doesn't actually need the entire document to service a request. Likewise, often queries are sent to MongoDB with a projection, excluding substantial (in terms of byte size) parts of the document.
MongoDB has no ability to optimise these projections (except for covered queries which are not always practical).
MongoDB should provide the ability to specify a "split" in a document such that the query planner can optimise disk access such that only the fields included by a projection are accessed.
This would improve both write and read performance by allowing updates to subsets of the document to be targeted to specific disk regions. Likewise, queries which exclude large portions of the document can be optimised as they will no longer need to read potentially large sections of the disk.