This is likely just because the snapshot revision is taken into account when we figure out to where to dispatch the request, so every 5s by default, it will have the opportunity to hash to a different node. You should diversify your load test data to make sure you're doing a more representative set of requests