diff --git a/codex/blockexchange/engine/engine.nim b/codex/blockexchange/engine/engine.nim
index 17a5d820..3a2d8218 100644
--- a/codex/blockexchange/engine/engine.nim
+++ b/codex/blockexchange/engine/engine.nim
@@ -175,9 +175,14 @@ proc refreshBlockKnowledge(
     self: BlockExcEngine, peer: BlockExcPeerCtx
 ) {.async: (raises: [CancelledError]).} =
   if self.pendingBlocks.wantListLen > 0:
-    let cids = toSeq(self.pendingBlocks.wantList)
-    trace "Sending our want list to a peer", peer = peer.id, length = cids.len
-    await self.network.request.sendWantList(peer.id, cids, full = true)
+    # We send only blocks that the peer hasn't already told us that they already have.
+    let
+      peerHave = peer.peerHave
+      toAsk = self.pendingBlocks.wantList.toSeq.filterIt(it notin peerHave)
+
+    if toAsk.len > 0:
+      trace "Sending want list to a peer", peer = peer.id, length = toAsk.len
+      await self.network.request.sendWantList(peer.id, toAsk, full = true)
 
 proc refreshBlockKnowledge(self: BlockExcEngine) {.async: (raises: [CancelledError]).} =
   for peer in self.peers.peers.values.toSeq:
@@ -189,15 +194,13 @@ proc refreshBlockKnowledge(self: BlockExcEngine) {.async: (raises: [CancelledErr
     # want list in the coarsest way possible instead of over many
     # small updates.
     #
+    if peer.refreshInProgress:
+      trace "Peer refresh in progress", peer = peer.id
+      continue
+
     # In dynamic swarms, staleness will dominate latency.
     if peer.lastRefresh < self.pendingBlocks.lastInclusion or peer.isKnowledgeStale:
-      # FIXME: we update the lastRefresh before actually refreshing because otherwise
-      #   a slow peer will be bombarded with requests. If the request does fail or the
-      #   peer does not reply, a retrying block will eventually issue this again. This
-      #   is a complex and convoluted flow - ideally we should simply be tracking this
-      #   request and retrying it on the absence of a response, eventually disconnecting
-      #   the peer if it consistently fails to respond.
-      peer.refreshed()
+      peer.refreshRequested()
       # TODO: optimize this by keeping track of what was sent and sending deltas.
       #   This should allow us to run much more frequent refreshes, and be way more
       #   efficient about it.
@@ -399,6 +402,8 @@ proc blockPresenceHandler*(
   if peerCtx.isNil:
     return
 
+  peerCtx.refreshReplied()
+
   for blk in blocks:
     if presence =? Presence.init(blk):
       peerCtx.setPresence(presence)
diff --git a/codex/blockexchange/peers/peercontext.nim b/codex/blockexchange/peers/peercontext.nim
index c917b7ee..0e0d1060 100644
--- a/codex/blockexchange/peers/peercontext.nim
+++ b/codex/blockexchange/peers/peercontext.nim
@@ -25,12 +25,18 @@ import ../../logutils
 
 export payments, nitro
 
+const
+  MinRefreshInterval = 5.seconds
+  MaxRefreshBackoff = 36 # 3 minutes
+
 type BlockExcPeerCtx* = ref object of RootObj
   id*: PeerId
   blocks*: Table[BlockAddress, Presence] # remote peer have list including price
   wantedBlocks*: HashSet[BlockAddress] # blocks that the peer wants
   exchanged*: int # times peer has exchanged with us
+  refreshInProgress*: bool # indicates if a refresh is in progress
   lastRefresh*: Moment # last time we refreshed our knowledge of the blocks this peer has
+  refreshBackoff*: int = 1 # backoff factor for refresh requests
   account*: ?Account # ethereum account of this peer
   paymentChannel*: ?ChannelId # payment channel id
   blocksSent*: HashSet[BlockAddress] # blocks sent to peer
@@ -39,7 +45,7 @@ type BlockExcPeerCtx* = ref object of RootObj
   activityTimeout*: Duration
 
 proc isKnowledgeStale*(self: BlockExcPeerCtx): bool =
-  self.lastRefresh + 5.minutes < Moment.now()
+  self.lastRefresh + self.refreshBackoff * MinRefreshInterval < Moment.now()
 
 proc isBlockSent*(self: BlockExcPeerCtx, address: BlockAddress): bool =
   address in self.blocksSent
@@ -50,9 +56,19 @@ proc markBlockAsSent*(self: BlockExcPeerCtx, address: BlockAddress) =
 proc markBlockAsNotSent*(self: BlockExcPeerCtx, address: BlockAddress) =
   self.blocksSent.excl(address)
 
-proc refreshed*(self: BlockExcPeerCtx) =
+proc refreshRequested*(self: BlockExcPeerCtx) =
+  trace "Refresh requested for peer", peer = self.id, backoff = self.refreshBackoff
+  self.refreshInProgress = true
   self.lastRefresh = Moment.now()
 
+proc refreshReplied*(self: BlockExcPeerCtx) =
+  self.refreshInProgress = false
+  self.lastRefresh = Moment.now()
+  self.refreshBackoff = min(self.refreshBackoff * 2, MaxRefreshBackoff)
+
+proc havesUpdated(self: BlockExcPeerCtx) =
+  self.refreshBackoff = 1
+
 proc peerHave*(self: BlockExcPeerCtx): HashSet[BlockAddress] =
   # XXX: this is ugly an inefficient, but since those will typically
   #  be used in "joins", it's better to pay the price here and have
@@ -63,6 +79,9 @@ proc contains*(self: BlockExcPeerCtx, address: BlockAddress): bool =
   address in self.blocks
 
 func setPresence*(self: BlockExcPeerCtx, presence: Presence) =
+  if presence.address notin self.blocks:
+    self.havesUpdated()
+
   self.blocks[presence.address] = presence
 
 func cleanPresence*(self: BlockExcPeerCtx, addresses: seq[BlockAddress]) =