SolrCloud cluster has a long latency for results to be available for search with NRT

Question

I'm not an expert on solr or lucene, but I'm configuring a SolrCloud cluster for NRT as we are moving away from WebSolr.

We have 2 environments: staging and production, staging works fantastically well,

Staging it's 1 zookeeper node and 2 solr nodes with several collections, 2 shards per collection with 2 as replication factor. Indexing and search is super fast.

Production it's a 5 zookeeper ensemble, with 3 solr nodes, only 1 collection with 2 shards and replication factor of 3.

AutoCommit is set for 60000 with a false open searcher and AutoSoftCommit is set to 1000.

Both environments have the same config. and both have a load balancer in front of our app.

On a brand new collection, NRT is fast, results are available within maybe 1s, but once we do a full index (9M records), updates become available for search after 2:30 MINUTES.

I can't figure out what I'm doing wrong, but I'm assuming it's due to the size of the cluster.

Any help will be highly appreciated.

Update: Additional note, I'm using 8gb ram instances on EC2 and my heap is at 4gb. Not sure if I need more for the volume of data I'm dealing with, since on our staging environment it works very fast.