diff --git a/test/e2e_node/node_perf_test.go b/test/e2e_node/node_perf_test.go
index e0b2089e795..81e38a44250 100644
--- a/test/e2e_node/node_perf_test.go
+++ b/test/e2e_node/node_perf_test.go
@@ -81,6 +81,15 @@ var _ = SIGDescribe("Node Performance Testing [Serial] [Slow] [Flaky]", func() {
 			GracePeriodSeconds: &gp,
 		}
 		f.PodClient().DeleteSync(pod.Name, delOpts, framework.DefaultPodDeletionTimeout)
+
+		// We are going to give some more time for the CPU manager to do any clean
+		// up it needs to do now that the pod has been deleted. Otherwise we may
+		// run into a data race condition in which the PostTestExec function
+		// deletes the CPU manager's checkpoint file while the CPU manager is still
+		// doing work and we end with a new checkpoint file after PosttestExec has
+		// finished. This issues would result in the kubelet panicking after we try
+		// and set the kubelet config.
+		time.Sleep(15 * time.Second)
 		ginkgo.By("running the post test exec from the workload")
 		err := wl.PostTestExec()
 		framework.ExpectNoError(err)