diff --git a/node/Constants.hpp b/node/Constants.hpp
index 5e052bcf9..83c98e0fc 100644
--- a/node/Constants.hpp
+++ b/node/Constants.hpp
@@ -250,9 +250,9 @@
 #define ZT_MULTICAST_TRANSMIT_TIMEOUT 5000
 
 /**
- * Default maximum number of peers to address with a single multicast (if unspecified in network)
+ * Default maximum number of peers to address with a single multicast (if unspecified in network config)
  */
-#define ZT_MULTICAST_DEFAULT_LIMIT 64
+#define ZT_MULTICAST_DEFAULT_LIMIT 32
 
 /**
  * Delay between scans of the topology active peer DB for peers that need ping
diff --git a/testnet/README.md b/testnet/README.md
index 6e1289328..3d3b3b54c 100644
--- a/testnet/README.md
+++ b/testnet/README.md
@@ -27,7 +27,7 @@ That will do a unicast all-to-all test and report results. At first latencies mi
 
     multicast <some node's 10-digit ZT address> * ffffffffffffffff 24 60
 
-This will send a multicast packet to ff:ff:ff:ff:ff:ff (broadcast) and report back who receives it. You should see multicast propagation limited to 64 nodes, since this is the setting for multicast limit on the fake test network (and the default if not overridden in netconf). Multicast will show the same "warm up" behavior as unicast.
+This will send a multicast packet to ff:ff:ff:ff:ff:ff (broadcast) and report back who receives it. You should see multicast propagation limited to 32 nodes, since this is the setting for multicast limit on the fake test network (and the default if not overridden in netconf). Multicast will show the same "warm up" behavior as unicast.
 
 Typing just "." will execute the same testnet command again.