Hello, it sounds like you're seeing a GC or queue issue. In the TSD there is up to a 1 second delay by default as data is batched per region server to reduce the calls to HBase. If the TSD is GCin'g then you could see an increased delay. Enable Java GC logs and see if anything looks suspicious.
Also, HBase may be GCing as well so enable GC logging on the region servers as well and see if anything is happening there. And there is an RPC queue in HBase that you can monitor. If you see a large queue size or huge rpc processing latencies without GC, then you'll want to see if you have any disk or IO issues.