diff --git a/src/plugins/nat/nat.c b/src/plugins/nat/nat.c
index 0ca2042..91f14d6 100644
--- a/src/plugins/nat/nat.c
+++ b/src/plugins/nat/nat.c
@@ -4752,6 +4752,7 @@
     [NAT_NEXT_IN2OUT_ED_OUTPUT_SLOW_PATH] = "nat44-ed-in2out-output-slowpath",
     [NAT_NEXT_OUT2IN_ED_FAST_PATH] = "nat44-ed-out2in",
     [NAT_NEXT_OUT2IN_ED_SLOW_PATH] = "nat44-ed-out2in-slowpath",
+    [NAT_NEXT_OUT2IN_ED_HANDOFF] = "nat44-ed-out2in-handoff",
     [NAT_NEXT_IN2OUT_CLASSIFY] = "nat44-in2out-worker-handoff",
     [NAT_NEXT_OUT2IN_CLASSIFY] = "nat44-out2in-worker-handoff",
   },
diff --git a/src/plugins/nat/nat.h b/src/plugins/nat/nat.h
index a713f7b..daebe45 100644
--- a/src/plugins/nat/nat.h
+++ b/src/plugins/nat/nat.h
@@ -88,6 +88,7 @@
   NAT_NEXT_IN2OUT_ED_OUTPUT_SLOW_PATH,
   NAT_NEXT_OUT2IN_ED_FAST_PATH,
   NAT_NEXT_OUT2IN_ED_SLOW_PATH,
+  NAT_NEXT_OUT2IN_ED_HANDOFF,
   NAT_NEXT_IN2OUT_CLASSIFY,
   NAT_NEXT_OUT2IN_CLASSIFY,
   NAT_N_NEXT,
@@ -162,6 +163,19 @@
     NAT_IN2OUT_ED_N_ERROR,
 } nat_in2out_ed_error_t;
 
+#define foreach_nat44_handoff_error                       \
+_(CONGESTION_DROP, "congestion drop")                     \
+_(SAME_WORKER, "same worker")                             \
+_(DO_HANDOFF, "do handoff")
+
+typedef enum
+{
+#define _(sym,str) NAT44_HANDOFF_ERROR_##sym,
+  foreach_nat44_handoff_error
+#undef _
+    NAT44_HANDOFF_N_ERROR,
+} nat44_handoff_error_t;
+
 #define foreach_nat_out2in_ed_error                     \
 _(UNSUPPORTED_PROTOCOL, "unsupported protocol")         \
 _(OUT_OF_PORTS, "out of ports")                         \
diff --git a/src/plugins/nat/nat44_handoff.c b/src/plugins/nat/nat44_handoff.c
index 69cdc42..651c8d1 100644
--- a/src/plugins/nat/nat44_handoff.c
+++ b/src/plugins/nat/nat44_handoff.c
@@ -33,26 +33,12 @@
   u8 output;
 } nat44_handoff_trace_t;
 
-#define foreach_nat44_handoff_error                       \
-_(CONGESTION_DROP, "congestion drop")                     \
-_(SAME_WORKER, "same worker")                             \
-_(DO_HANDOFF, "do handoff")
-
-typedef enum
-{
-#define _(sym,str) NAT44_HANDOFF_ERROR_##sym,
-  foreach_nat44_handoff_error
-#undef _
-    NAT44_HANDOFF_N_ERROR,
-} nat44_handoff_error_t;
-
 static char *nat44_handoff_error_strings[] = {
 #define _(sym,string) string,
   foreach_nat44_handoff_error
 #undef _
 };
 
-
 static u8 *
 format_nat44_handoff_trace (u8 * s, va_list * args)
 {
diff --git a/src/plugins/nat/nat_ha.c b/src/plugins/nat/nat_ha.c
index 3ae572b..08b065d 100644
--- a/src/plugins/nat/nat_ha.c
+++ b/src/plugins/nat/nat_ha.c
@@ -1057,7 +1057,7 @@
 #define _(sym,str) NAT_HA_HANDOFF_ERROR_##sym,
   foreach_nat_ha_handoff_error
 #undef _
-    NAT44_HANDOFF_N_ERROR,
+    NAT_HA_HANDOFF_N_ERROR,
 } nat_ha_handoff_error_t;
 
 static char *nat_ha_handoff_error_strings[] = {
diff --git a/src/plugins/nat/out2in_ed.c b/src/plugins/nat/out2in_ed.c
index b0e8892..95e2ea9 100644
--- a/src/plugins/nat/out2in_ed.c
+++ b/src/plugins/nat/out2in_ed.c
@@ -46,6 +46,11 @@
   u32 is_slow_path;
 } nat44_ed_out2in_trace_t;
 
+typedef struct
+{
+  u16 thread_next;
+} nat44_ed_out2in_handoff_trace_t;
+
 static u8 *
 format_nat44_ed_out2in_trace (u8 * s, va_list * args)
 {
@@ -513,16 +518,24 @@
   nat44_session_update_lru (sm, s, thread_index);
 }
 
-static inline void
-create_bypass_for_fwd_worker (snat_main_t * sm, vlib_buffer_t * b,
-			      ip4_header_t * ip, u32 rx_fib_index)
+static_always_inline int
+create_bypass_for_fwd_worker (snat_main_t * sm,
+			      vlib_buffer_t * b, ip4_header_t * ip,
+			      u32 rx_fib_index, u32 thread_index)
 {
-  ip4_header_t ip_wkr = {
+  ip4_header_t tmp = {
     .src_address = ip->dst_address,
   };
-  u32 thread_index = sm->worker_in2out_cb (&ip_wkr, rx_fib_index, 0);
+  u32 index = sm->worker_in2out_cb (&tmp, rx_fib_index, 0);
+
+  if (index != thread_index)
+    {
+      vnet_buffer2 (b)->nat.thread_next = index;
+      return 1;
+    }
 
   create_bypass_for_fwd (sm, b, ip, rx_fib_index, thread_index);
+  return 0;
 }
 
 #ifndef CLIB_MARCH_VARIANT
@@ -590,10 +603,19 @@
 	      else
 		{
 		  if (sm->num_workers > 1)
-		    create_bypass_for_fwd_worker (sm, b, ip, rx_fib_index);
+		    {
+		      if (create_bypass_for_fwd_worker (sm, b, ip,
+							rx_fib_index,
+							thread_index))
+			{
+			  next = NAT_NEXT_OUT2IN_ED_HANDOFF;
+			}
+		    }
 		  else
-		    create_bypass_for_fwd (sm, b, ip, rx_fib_index,
-					   thread_index);
+		    {
+		      create_bypass_for_fwd (sm, b, ip, rx_fib_index,
+					     thread_index);
+		    }
 		}
 	    }
 	  goto out;
@@ -1208,12 +1230,18 @@
 		    }
 		  else
 		    {
-		      if (sm->num_workers > 1)
-			create_bypass_for_fwd_worker (sm, b0, ip0,
-						      rx_fib_index0);
+		      if ((sm->num_workers > 1)
+			  && create_bypass_for_fwd_worker (sm, b0, ip0,
+							   rx_fib_index0,
+							   thread_index))
+			{
+			  next[0] = NAT_NEXT_OUT2IN_ED_HANDOFF;
+			}
 		      else
-			create_bypass_for_fwd (sm, b0, ip0, rx_fib_index0,
-					       thread_index);
+			{
+			  create_bypass_for_fwd (sm, b0, ip0, rx_fib_index0,
+						 thread_index);
+			}
 		    }
 		}
 	      goto trace0;
@@ -1383,6 +1411,88 @@
   return frame->n_vectors;
 }
 
+static inline uword
+nat_handoff_node_fn_inline (vlib_main_t * vm,
+			    vlib_node_runtime_t * node,
+			    vlib_frame_t * frame, u32 fq_index)
+{
+  u32 n_enq, n_left_from, *from;
+
+  u16 thread_indices[VLIB_FRAME_SIZE], *ti = thread_indices;
+  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b = bufs;
+
+  from = vlib_frame_vector_args (frame);
+  n_left_from = frame->n_vectors;
+
+  vlib_get_buffers (vm, from, b, n_left_from);
+
+  while (n_left_from >= 4)
+    {
+      if (PREDICT_TRUE (n_left_from >= 8))
+	{
+	  vlib_prefetch_buffer_header (b[4], LOAD);
+	  vlib_prefetch_buffer_header (b[5], LOAD);
+	  vlib_prefetch_buffer_header (b[6], LOAD);
+	  vlib_prefetch_buffer_header (b[7], LOAD);
+	  CLIB_PREFETCH (&b[4]->data, CLIB_CACHE_LINE_BYTES, LOAD);
+	  CLIB_PREFETCH (&b[5]->data, CLIB_CACHE_LINE_BYTES, LOAD);
+	  CLIB_PREFETCH (&b[6]->data, CLIB_CACHE_LINE_BYTES, LOAD);
+	  CLIB_PREFETCH (&b[7]->data, CLIB_CACHE_LINE_BYTES, LOAD);
+	}
+
+      ti[0] = vnet_buffer2 (b[0])->nat.thread_next;
+      ti[1] = vnet_buffer2 (b[1])->nat.thread_next;
+      ti[2] = vnet_buffer2 (b[2])->nat.thread_next;
+      ti[3] = vnet_buffer2 (b[3])->nat.thread_next;
+
+      b += 4;
+      ti += 4;
+      n_left_from -= 4;
+    }
+
+  while (n_left_from > 0)
+    {
+      ti[0] = vnet_buffer2 (b[0])->nat.thread_next;
+
+      b += 1;
+      ti += 1;
+      n_left_from -= 1;
+    }
+
+  if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE)))
+    {
+      u32 i;
+      b = bufs;
+      ti = thread_indices;
+
+      for (i = 0; i < frame->n_vectors; i++)
+	{
+	  if (b[0]->flags & VLIB_BUFFER_IS_TRACED)
+	    {
+	      nat44_ed_out2in_handoff_trace_t *t =
+		vlib_add_trace (vm, node, b[0], sizeof (*t));
+	      t->thread_next = ti[0];
+	      b += 1;
+	      ti += 1;
+	    }
+	  else
+	    break;
+	}
+    }
+
+  n_enq = vlib_buffer_enqueue_to_thread (vm, fq_index, from, thread_indices,
+					 frame->n_vectors, 1);
+
+  if (n_enq < frame->n_vectors)
+    {
+      vlib_node_increment_counter (vm, node->node_index,
+				   NAT44_HANDOFF_ERROR_CONGESTION_DROP,
+				   frame->n_vectors - n_enq);
+    }
+
+  return frame->n_vectors;
+}
+
 VLIB_NODE_FN (nat44_ed_out2in_node) (vlib_main_t * vm,
 				     vlib_node_runtime_t * node,
 				     vlib_frame_t * frame)
@@ -1431,6 +1541,35 @@
 /* *INDENT-ON* */
 
 static u8 *
+format_nat44_ed_out2in_handoff_trace (u8 * s, va_list * args)
+{
+  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
+  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
+  nat44_ed_out2in_handoff_trace_t *t =
+    va_arg (*args, nat44_ed_out2in_handoff_trace_t *);
+  return format (s, "out2in ed handoff thread_next index %d", t->thread_next);
+}
+
+VLIB_NODE_FN (nat44_ed_out2in_handoff_node) (vlib_main_t * vm,
+					     vlib_node_runtime_t * node,
+					     vlib_frame_t * frame)
+{
+  return nat_handoff_node_fn_inline (vm, node, frame,
+				     snat_main.ed_out2in_node_index);
+}
+
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (nat44_ed_out2in_handoff_node) = {
+  .name = "nat44-ed-out2in-handoff",
+  .vector_size = sizeof (u32),
+  .sibling_of = "nat-default",
+  .format_trace = format_nat44_ed_out2in_handoff_trace,
+  .type = VLIB_NODE_TYPE_INTERNAL,
+  .n_errors = 0,
+};
+/* *INDENT-ON* */
+
+static u8 *
 format_nat_pre_trace (u8 * s, va_list * args)
 {
   CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
diff --git a/src/vnet/buffer.h b/src/vnet/buffer.h
index 50515c4..9e997b8 100644
--- a/src/vnet/buffer.h
+++ b/src/vnet/buffer.h
@@ -452,11 +452,10 @@
   /* size of L4 prototol header */
   u16 gso_l4_hdr_sz;
 
-  /* The union below has a u64 alignment, so this space is unused */
-  u32 __unused2[1];
-
   struct
   {
+    u16 unused;
+    u16 thread_next;
     u32 arc_next;
     u32 ed_out2in_nat_session_index;
   } nat;