The GridFS is a great feature to the MongoDB.
I'm thinking what the performance affected by random reading /writing to a dynamic GridFS file. The chunk data is huge and it looks like only the whole chunk can be updated to the GridFS file which at some time would require lot of data streaming for random writing. Here is the idea to add support paging within the chunk data as for that chunk data is mainly for storage while the paging is for client caching and updating, so that we can keep large size for the chunk data for storage and paging can split the chunk data into small subsets for client caching/updating. I think this will require a lot of work to be able to split the chunk in client and server needs to allow in place updating partial chunk data.
I'm not sure if similar feature is available or will be added in future?