diff --git a/test/e2e/helpers_test.go b/test/e2e/helpers_test.go
index 83568b79e8..850e14d187 100644
--- a/test/e2e/helpers_test.go
+++ b/test/e2e/helpers_test.go
@@ -454,6 +454,12 @@ func deployIngressWithEchoBackends(ctx context.Context, t *testing.T, env enviro
 
 	t.Logf("exposing deployment %s via service", deployment.Name)
 	service := generators.NewServiceForDeployment(deployment, corev1.ServiceTypeClusterIP)
+	for i := range service.Spec.Ports {
+		// Set Service's appProtocol to http so that Kuma load-balances requests on HTTP-level instead of TCP.
+		// TCP load-balancing proved to cause issues with not distributing load evenly in short time spans like in this test.
+		// Ref: https://github.com/Kong/kubernetes-ingress-controller/issues/5498
+		service.Spec.Ports[i].AppProtocol = lo.ToPtr("http")
+	}
 	_, err = env.Cluster().Client().CoreV1().Services(corev1.NamespaceDefault).Create(ctx, service, metav1.CreateOptions{})
 	require.NoError(t, err)
 	t.Cleanup(func() {