I found this great forum thread about tuning index buffer size and how it is distributed amongs active shards. This suggests that you don't have to optimize this setting for the index that is beeing written, it will just use the buffer for shards that are active (i.e. beeing written to).
Another thing to be aware of for your use case might be the shrinking api that is coming up as a new feature in the soon-to-be-releases 5.0 version of Elasticsearch. This will allow you to basically use a higher shard number for the indices that are actively written to while reducing the shard count for the ones that are only read from. This blog post explains it in a little bit more detail, but mind you this is a feature thats not available in the 2.x versions of ES yet.