According to what I've seen on the AWS forums, this issue is more apparent on "older" generation instances. I'm currently running on an m3.medium which is somewhat old as the latest generation is m5.
The currently recommended node by the aws marketplace is c4.large. This is a larger machine than what I have now, but it might not experience the same issue.
Is anyone seeing the same issue with the c4 generation nodes? If the marketplace would allow, we could try it out on a more recent generation node just to see what the difference would be.