Settings for improve latency in multithreading requests

I'm doing some performance test for future use ES in server application

I'm start ES with stock settings (one additional only - memory). I have 2 indexes (in feature will be 25) with 5 shards in each. Each index have two types of documents. First type contains 20-120 documents and updates every 3s, second type contains 1500-3200 documents and updates every 10min.
Each document have Average document size - 12kb. I use ES for store and search documents.
ES starts in AWS c1.xlarge instance 8cores, 7GB RAM.

I use ApacheBenchmark to test performance. I noticed that with increasing concurrency latency is rising.
similar POST request

Measurement results of one response latency

response size: 729b
1000 requests in 1 thread - 5ms
1000 requests in 100 threads - 53ms

response size: 573kb
1000 requests in 1 thread - 24ms
1000 requests in 100 threads - 5520ms

What I can tune to improve latency.

Sorry for my English.