narf, index.routing.allocation.total_shards_per_node is what I was looking for, please disregard this post (firefighting blindness)
We're seeing issues with the auto rebalancing of shards, I/O expensive shards tend to cluster on one or two nodes, moving inexpensive shards of other indices to the rest of the cluster. The result: extreme load and unresponsive services.
Is there a hidden setting to only allow x number of shards for a specific I/O intensive index? (e.g. we have 75 primary shards and 35 servers, therefore only allowing a maximum of 4 shards per node for that index would be a safeguard against the clustering-of-death)