diff --git a/test/e2e/autoscaling/cluster_size_autoscaling.go b/test/e2e/autoscaling/cluster_size_autoscaling.go
index 9eb0ca5f521..6790f895603 100644
--- a/test/e2e/autoscaling/cluster_size_autoscaling.go
+++ b/test/e2e/autoscaling/cluster_size_autoscaling.go
@@ -1007,7 +1007,8 @@ var _ = SIGDescribe("Cluster size autoscaling", framework.WithSlow(), func() {
 		// 70% of allocatable memory of a single node * replica count, forcing a scale up in case of normal pods
 		replicaCount := 2 * nodeCount
 		reservedMemory := int(float64(replicaCount) * float64(0.7) * float64(memAllocatableMb))
-		ginkgo.DeferCleanup(ReserveMemoryWithSchedulerName(ctx, f, "memory-reservation", replicaCount, reservedMemory, false, 1, nonExistingBypassedSchedulerName))
+		cleanupFunc := ReserveMemoryWithSchedulerName(ctx, f, "memory-reservation", replicaCount, reservedMemory, false, 1, nonExistingBypassedSchedulerName)
+		defer cleanupFunc()
 		// Verify that cluster size is increased
 		ginkgo.By("Waiting for cluster scale-up")
 		sizeFunc := func(size int) bool {
@@ -1020,7 +1021,8 @@ var _ = SIGDescribe("Cluster size autoscaling", framework.WithSlow(), func() {
 		// 50% of allocatable memory of a single node, so that no scale up would trigger in normal cases
 		replicaCount := 1
 		reservedMemory := int(float64(0.5) * float64(memAllocatableMb))
-		ginkgo.DeferCleanup(ReserveMemoryWithSchedulerName(ctx, f, "memory-reservation", replicaCount, reservedMemory, false, 1, nonExistingBypassedSchedulerName))
+		cleanupFunc := ReserveMemoryWithSchedulerName(ctx, f, "memory-reservation", replicaCount, reservedMemory, false, 1, nonExistingBypassedSchedulerName)
+		defer cleanupFunc()
 		// Verify that cluster size is the same
 		ginkgo.By(fmt.Sprintf("Waiting for scale up hoping it won't happen, polling cluster size for %s", scaleUpTimeout.String()))
 		sizeFunc := func(size int) bool {
@@ -1035,7 +1037,8 @@ var _ = SIGDescribe("Cluster size autoscaling", framework.WithSlow(), func() {
 		replicaCount := 2 * nodeCount
 		reservedMemory := int(float64(replicaCount) * float64(0.7) * float64(memAllocatableMb))
 		schedulerName := "non-existent-scheduler-" + f.UniqueName
-		ginkgo.DeferCleanup(ReserveMemoryWithSchedulerName(ctx, f, "memory-reservation", replicaCount, reservedMemory, false, 1, schedulerName))
+		cleanupFunc := ReserveMemoryWithSchedulerName(ctx, f, "memory-reservation", replicaCount, reservedMemory, false, 1, schedulerName)
+		defer cleanupFunc()
 		// Verify that cluster size is the same
 		ginkgo.By(fmt.Sprintf("Waiting for scale up hoping it won't happen, polling cluster size for %s", scaleUpTimeout.String()))
 		sizeFunc := func(size int) bool {