diff --git a/zephyr/lib/actions.py b/zephyr/lib/actions.py
index d7d3d582ca..8a5280f485 100644
--- a/zephyr/lib/actions.py
+++ b/zephyr/lib/actions.py
@@ -150,14 +150,14 @@ def do_send_message(message, no_log=False):
 
     # We can only publish messages to longpolling clients if the Tornado server is running.
     if settings.TORNADO_SERVER:
-        # Render Markdown etc. here, so that the single-threaded Tornado server doesn't have to.
-        # TODO: Reduce duplication in what we send.
-        rendered = { 'text/html':       message.to_dict(apply_markdown=True),
-                     'text/x-markdown': message.to_dict(apply_markdown=False) }
+        # Render Markdown etc. here and store (automatically) in
+        # memcached, so that the single-threaded Tornado server
+        # doesn't have to.
+        message.to_dict(apply_markdown=True)
+        message.to_dict(apply_markdown=False)
         requests.post(settings.TORNADO_SERVER + '/notify_new_message', data=dict(
             secret   = settings.SHARED_SECRET,
             message  = message.id,
-            rendered = simplejson.dumps(rendered),
             users    = simplejson.dumps([str(user.id) for user in recipients])))
 
 def create_stream_if_needed(realm, stream_name):
diff --git a/zephyr/models.py b/zephyr/models.py
index 55d5760557..addf60605d 100644
--- a/zephyr/models.py
+++ b/zephyr/models.py
@@ -171,14 +171,6 @@ class Message(models.Model):
 
     @cache_with_key(lambda self, apply_markdown: 'message_dict:%d:%d' % (self.id, apply_markdown))
     def to_dict(self, apply_markdown):
-        # Messages arrive in the Tornado process with the dicts already rendered.
-        # This avoids running the Markdown parser and some database queries in the single-threaded
-        # Tornado server.
-        #
-        # This field is not persisted to the database and will disappear if the object is re-fetched.
-        if hasattr(self, 'precomputed_dicts'):
-            return self.precomputed_dicts['text/html' if apply_markdown else 'text/x-markdown']
-
         display_recipient = get_display_recipient(self.recipient)
         if self.recipient.type == Recipient.STREAM:
             display_type = "stream"
diff --git a/zephyr/tornadoviews.py b/zephyr/tornadoviews.py
index ad4ae9c2f7..1ee3441727 100644
--- a/zephyr/tornadoviews.py
+++ b/zephyr/tornadoviews.py
@@ -139,13 +139,6 @@ def notify_new_message(request):
     recipient_profile_ids = map(int, json_to_list(request.POST['users']))
     message = cache_get_message(int(request.POST['message']))
 
-    # Cause message.to_dict() to return the dicts already rendered in the other process.
-    #
-    # We decode this JSON only to eventually re-encode it as JSON.
-    # This isn't trivial to fix, because we do access some fields in the meantime
-    # (see send_with_safety_check).  It's probably not a big deal.
-    message.precomputed_dicts = simplejson.loads(request.POST['rendered'])
-
     for user_profile_id in recipient_profile_ids:
         receive(user_profile_id, message)