The saving process already converts data into json and performs successive bulkloading to Elasticsearch, so doing this outside of the connector probably won't see any changes to write speed. Ten-plus terrabytes seems like a large amount of data for a 5 node cluster, though I'm not familiar with your deployment so it's hard for me to say it's unreasonable.
One thing that might make sense to take a look at is the batch size settings on ES-Hadoop. Increasing these may allow you to send larger bulk requests and cut down on high amounts of back and forth traffic.
Additionally, disabling refresh on your indices and in ES-Hadoop may help the ingestion rate in that Elasticsearch may spend less time flushing smaller segments to disk.