diff --git a/qa/L0_http/http_request_many_chunks.py b/qa/L0_http/http_request_many_chunks.py
index d733fb000c..aef2ed97b1 100755
--- a/qa/L0_http/http_request_many_chunks.py
+++ b/qa/L0_http/http_request_many_chunks.py
@@ -30,7 +30,11 @@
 import unittest
 
 sys.path.append("../common")
-from test_util import MIB, get_server_process_from_env, wait_for_stable_rss
+from test_util import (  # noqa: E402
+    MIB,
+    get_server_process_from_env,
+    wait_for_stable_rss,
+)
 
 
 class HTTPRequestManyChunksTest(unittest.TestCase):
@@ -142,7 +146,7 @@ def test_chunked_infer_over_max_chunks_reject_with_bounded_rss_growth(self):
         )
         # Wait until RSS is stable across several measurements before continuing.
         server = get_server_process_from_env("SERVER_PID")
-        wait_for_stable_rss(server)
+        wait_for_stable_rss(server, stable_threshold=100)
 
         # Monitor RSS growth over 100 requests.
         repeat_request_count = 100
diff --git a/src/http_server.cc b/src/http_server.cc
index bd7262ed80..80f3c2edfd 100644
--- a/src/http_server.cc
+++ b/src/http_server.cc
@@ -281,6 +281,10 @@ HTTPServer::StopCallback(evutil_socket_t sock, short events, void* arg)
 void
 HTTPServer::Dispatch(evhtp_request_t* req, void* arg)
 {
+  // A parser hook may have replied before the buffered request body finishes.
+  if ((req->flags & EVHTP_REQ_FLAG_FINISHED) != 0) {
+    return;
+  }
   (static_cast<HTTPServer*>(arg))->Handle(req);
 }
 
@@ -317,6 +321,13 @@ HTTPServer::ChunkCountIncrement(
     return EVHTP_RES_OK;
   }
   if (++req->chunk_count > kMaxChunkedChunks) {
+    // Release the accumulated body before allocating the error response. The
+    // parser can still consume bytes from its current read buffer after reads
+    // are disabled, so discard those body fragments instead of retaining them.
+    evbuffer_drain(req->buffer_in, evbuffer_get_length(req->buffer_in));
+    evhtp_request_set_hook(
+        req, evhtp_hook_on_read,
+        (evhtp_hook)(void*)HTTPServer::DiscardRequestBody, nullptr);
     AddContentTypeHeader(req, "application/json");
     const std::string msg =
         std::string("Chunked request body exceeds maximum of ") +
@@ -336,6 +347,16 @@ HTTPServer::ChunkCountIncrement(
   return EVHTP_RES_OK;
 }
 
+evhtp_res
+HTTPServer::DiscardRequestBody(
+    evhtp_request_t* req, evbuffer* buffer, void* arg)
+{
+  (void)req;
+  (void)arg;
+  evbuffer_drain(buffer, evbuffer_get_length(buffer));
+  return EVHTP_RES_OK;
+}
+
 evhtp_res
 HTTPServer::EndConnection(evhtp_connection_t* conn, void* arg)
 {
diff --git a/src/http_server.h b/src/http_server.h
index f45d07f4e7..c14afc8f21 100644
--- a/src/http_server.h
+++ b/src/http_server.h
@@ -108,6 +108,8 @@ class HTTPServer {
   // Transfer-Encoding: chunked — increment count per non-empty chunk (capped).
   static evhtp_res ChunkCountIncrement(
       evhtp_request_t* req, uint64_t chunk_len, void* arg);
+  static evhtp_res DiscardRequestBody(
+      evhtp_request_t* req, evbuffer* buffer, void* arg);
 
   static evhtp_res NewConnection(evhtp_connection_t* conn, void* arg);
   static evhtp_res EndConnection(evhtp_connection_t* conn, void* arg);