There currently is no limit on the size of records in in-memory storage, but it may change. Just be aware that getting a single very large record is inefficient, as it will be filling up the network buffers. It would be more performant to chop it up into several smaller records and using a batch-read to fetch the data.