diff --git a/test/e2e/density.go b/test/e2e/density.go
index acb92a79fd0..9455fd50156 100644
--- a/test/e2e/density.go
+++ b/test/e2e/density.go
@@ -86,6 +86,13 @@ var _ = Describe("Density", func() {
 		expectNoError(err)
 		minionCount = len(minions.Items)
 		Expect(minionCount).NotTo(BeZero())
+
+		// Terminating a namespace (deleting the remaining objects from it - which
+		// generally means events) can affect the current run. Thus we wait for all
+		// terminating namespace to be finally deleted before starting this test.
+		err = deleteTestingNS(c)
+		expectNoError(err)
+
 		nsForTesting, err := createTestingNS("density", c)
 		ns = nsForTesting.Name
 		expectNoError(err)
@@ -149,7 +156,7 @@ var _ = Describe("Density", func() {
 		// TODO: Reenable once we can measure latency only from a single test.
 		// TODO: Expose runLatencyTest as ginkgo flag.
 		{podsPerMinion: 3, skip: true, runLatencyTest: false, interval: 10 * time.Second},
-		{podsPerMinion: 30, skip: true, runLatencyTest: false, interval: 10 * time.Second},
+		{podsPerMinion: 30, skip: true, runLatencyTest: true, interval: 10 * time.Second},
 		// More than 30 pods per node is outside our v1.0 goals.
 		// We might want to enable those tests in the future.
 		{podsPerMinion: 50, skip: true, runLatencyTest: false, interval: 10 * time.Second},
@@ -158,7 +165,7 @@ var _ = Describe("Density", func() {
 
 	for _, testArg := range densityTests {
 		name := fmt.Sprintf("should allow starting %d pods per node", testArg.podsPerMinion)
-		if testArg.podsPerMinion <= 30 {
+		if testArg.podsPerMinion == 30 {
 			name = "[Performance suite] " + name
 		}
 		if testArg.skip {
diff --git a/test/e2e/load.go b/test/e2e/load.go
index dc3b30ac36d..68cbf331463 100644
--- a/test/e2e/load.go
+++ b/test/e2e/load.go
@@ -63,6 +63,13 @@ var _ = Describe("Load capacity", func() {
 		expectNoError(err)
 		nodeCount = len(nodes.Items)
 		Expect(nodeCount).NotTo(BeZero())
+
+		// Terminating a namespace (deleting the remaining objects from it - which
+		// generally means events) can affect the current run. Thus we wait for all
+		// terminating namespace to be finally deleted before starting this test.
+		err = deleteTestingNS(c)
+		expectNoError(err)
+
 		nsForTesting, err := createTestingNS("load", c)
 		ns = nsForTesting.Name
 		expectNoError(err)
diff --git a/test/e2e/util.go b/test/e2e/util.go
index 0b3f133d459..250d5284923 100644
--- a/test/e2e/util.go
+++ b/test/e2e/util.go
@@ -376,6 +376,32 @@ func createTestingNS(baseName string, c *client.Client) (*api.Namespace, error)
 	return got, nil
 }
 
+// deleteTestingNS checks whether all e2e based existing namespaces are in the Terminating state
+// and waits until they are finally deleted.
+func deleteTestingNS(c *client.Client) error {
+	Logf("Waiting for terminating namespaces to be deleted...")
+	for start := time.Now(); time.Since(start) < 30*time.Minute; time.Sleep(15 * time.Second) {
+		namespaces, err := c.Namespaces().List(labels.Everything(), fields.Everything())
+		if err != nil {
+			Logf("Listing namespaces failed: %v", err)
+			continue
+		}
+		terminating := 0
+		for _, ns := range namespaces.Items {
+			if strings.HasPrefix(ns.ObjectMeta.Name, "e2e-tests-") {
+				if ns.Status.Phase == api.NamespaceActive {
+					return fmt.Errorf("Namespace %s is active", ns)
+				}
+				terminating++
+			}
+		}
+		if terminating == 0 {
+			return nil
+		}
+	}
+	return fmt.Errorf("Waiting for terminating namespaces to be deleted timed out")
+}
+
 func waitForPodRunningInNamespace(c *client.Client, podName string, namespace string) error {
 	return waitForPodCondition(c, namespace, podName, "running", podStartTimeout, func(pod *api.Pod) (bool, error) {
 		if pod.Status.Phase == api.PodRunning {