Pod is handling all requests even when Keep-alive is set to False

Question

I'm working on a university project related to studying how Kubernetes scaling works but I have been facing an issue: I have a cluster with one master node (for control plane) and one edge node and I have enabled HPA on a deployment I created, named stress-app, below is the deployment and service YAML files, respectively:

stress-app-deployment.yaml:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: stress-app
spec:
  replicas: 1
  selector:
    matchLabels:
      app: stress-app
  template:
    metadata:
      labels:
        app: stress-app
    spec:
      containers:
        - name: stress-app
          image: annis99/stress-app:v0.2
          imagePullPolicy: Always
          ports:
            - containerPort: 8081
          resources:
            requests:
              cpu: 0.5
              memory: 100Mi

stress-app-service.yaml

apiVersion: v1
kind: Service
metadata:
  name: stress-app-service
spec:
  selector:
    app: stress-app
  ports:
    - protocol: TCP
      port: 8081
      targetPort: 8081
  type: LoadBalancer

The app is a simple API built with FastAPI with one endpoint '/workload/cpu', which when called create a fake CPU load using the stress-ng CLI tool.

The app works fine, however, when I try to load-test the cluster to see the HPA scaling, it scales up to 8 replicas sometimes, but only one replica of the pod handles most requests, causing high CPU spikes for the node hosting that replica. I have also checked that other replicas are running and not in a pending state.

I have noticed this when trying with JMeter, Locust and K6 load testing tools, by setting keep-alive to False, however, when I try with the browser (Google Chrome) by opening several tabs and requesting that endpoint simultaneously, more replicas are involved and process the requests resulting in a fairer distribution of the workload. Any ideas why this happens with sophisticated load-testing tools?