diff --git a/agent-node/src/agent_node/core/watcher.py b/agent-node/src/agent_node/core/watcher.py
index 69dd04a..5285f09 100644
--- a/agent-node/src/agent_node/core/watcher.py
+++ b/agent-node/src/agent_node/core/watcher.py
@@ -173,7 +173,7 @@
 
             self.last_sync[rel_path] = file_hash
             file_size = os.path.getsize(abs_path)
-            chunk_size = 1024 * 1024 # 1MB buffer for hashing/stream
+            chunk_size = 4 * 1024 * 1024 # 4MB buffer for hashing/stream
             total_chunks = (file_size + chunk_size - 1) // chunk_size if file_size > 0 else 1
             
             print(f"    [📁📤] Streaming Sync Started: {rel_path} ({file_size} bytes)")
diff --git a/agent-node/src/agent_node/node.py b/agent-node/src/agent_node/node.py
index 8b54593..d56b806 100644
--- a/agent-node/src/agent_node/node.py
+++ b/agent-node/src/agent_node/node.py
@@ -33,7 +33,8 @@
         self.sync_mgr = NodeSyncManager()
         self.skills = SkillManager(max_workers=config.MAX_SKILL_WORKERS, sync_mgr=self.sync_mgr)
         self.watcher = WorkspaceWatcher(self._on_sync_delta)
-        self.task_queue = queue.Queue(maxsize=10000) # Prevents fast PTY/Worker deadlock
+        # Bounded queue to prevent memory ballooning; 250 * 4MB chunks = 1GB max memory.
+        self.task_queue = queue.Queue(maxsize=250) 
         self.stub = None
         self.channel = None
         self._stop_event = threading.Event()
diff --git a/ai-hub/app/core/grpc/services/assistant.py b/ai-hub/app/core/grpc/services/assistant.py
index 5c9e8f9..bd4c443 100644
--- a/ai-hub/app/core/grpc/services/assistant.py
+++ b/ai-hub/app/core/grpc/services/assistant.py
@@ -212,13 +212,13 @@
                     )
                 ), priority=2)
 
-        # M6: Use registry executor if available for parallel mesh broadcast
-        if self.registry.executor:
-            for nid in destinations:
-                self.registry.executor.submit(_send_to_node, nid)
-        else:
-            for nid in destinations:
-                _send_to_node(nid)
+        # M6: We MUST execute synchronously here instead of using the registry executor.
+        # If we offload to an unbounded thread pool, the inbound sender will push chunks faster
+        # than the outbound connection can send them, queuing gigabytes of protobufs in memory
+        # and causing the Hub to OOM instantly on large files.
+        # Running synchronously provides natural backpressure to the sender queue.
+        for nid in destinations:
+            _send_to_node(nid)
 
     def broadcast_delete(self, session_id: str, sender_node_id: str, rel_path: str):
         """Broadcasts a file deletion from one node to all other nodes in the session mesh."""
diff --git a/ai-hub/app/core/services/node_registry.py b/ai-hub/app/core/services/node_registry.py
index 971fb02..49200e6 100644
--- a/ai-hub/app/core/services/node_registry.py
+++ b/ai-hub/app/core/services/node_registry.py
@@ -52,8 +52,8 @@
         self.node_id = node_id
         self.user_id = user_id           # Owner — maps node to a Hub user
         self.metadata = metadata         # desc, caps (capabilities dict)
-        # Increased queue size to 1000 to handle high-concurrency file sync without dropping interactive tasks
-        self.queue: queue.PriorityQueue = queue.PriorityQueue(maxsize=1000)
+        # Bounded queue to prevent memory ballooning; 100 * 4MB chunks = 400MB max memory per connected node
+        self.queue: queue.PriorityQueue = queue.PriorityQueue(maxsize=100)
         self.stats: dict = {
             "active_worker_count": 0,
             "cpu_usage_percent": 0.0,
@@ -75,7 +75,9 @@
         
         def _blocking_put():
             try:
-                self.queue.put(item, block=True, timeout=2.0)
+                # 300s timeout provides absolute maximum backpressure for large file sync across slow VPNs
+                # preventing dropped chunks while still protecting against permanent deadlocks.
+                self.queue.put(item, block=True, timeout=300.0)
             except queue.Full:
                 logger.warning(f"[📋⚠️] Message dropped for {self.node_id}: outbound queue FULL. Node may be unresponsive.")
             except Exception as e: