'delete noreply' may hang the client

I use libmemcached v0.30 in noreply mode. And find that bug: in case
of deleting of a large amount of keys from memcached my production
server start waiting for a while on read().
This behaviour is reproducing in the test environment each time but i
can't reproduce it in the libmemcached unit tests.
Version v0.31 ans 0.32 have the same behaviour.
If I use libmemcached in usual mode (noreply is off) this thing does
not happened.

(gdb) bt
#0 0x00007f0f9ca8758b in read () from /lib/libpthread.so.0
#1 0x00000000006808e3 in memcached_io_read (ptr=0x4e2b5720,
buffer=0x7f0f9bd73b60, length=1, nread=0x7f0f9bd73a80) at
memcached_io.c:108
#2 0x0000000000681490 in memcached_io_readline (ptr=0x4e2b5720,
buffer_ptr=0x7f0f9bd73b60 "\1", size=1024) at memcached_io.c:482
#3 0x000000000068330d in textual_read_one_response (ptr=0x4e2b5720,
buffer=0x7f0f9bd73b60 "\1", buffer_length=1024, result=0x7f0f9bd73f60)
   at memcached_response.c:208
#4 0x0000000000682ccb in memcached_read_one_response (ptr=0x4e2b5720,
buffer=0x7f0f9bd73b60 "\1", buffer_length=1024, result=0x7f0f9bd73f60)
   at memcached_response.c:31
#5 0x0000000000681841 in memcached_purge (ptr=0x4e2b5720) at
memcached_purge.c:56
#6 0x00000000006810a3 in io_flush (ptr=0x4e2b5720,
error=0x7f0f9bd74188) at memcached_io.c:340
#7 0x0000000000680c87 in memcached_io_write (ptr=0x4e2b5720,
buffer=0x7f0f9bd74270, length=0, with_flush=1 '\1') at
memcached_io.c:241
#8 0x000000000067e040 in memcached_do (ptr=0x4e2b5720,
command=0x7f0f9bd74270, command_length=28, with_flush=1 '\1') at
memcached_do.c:26
#9 0x000000000067dc7e in memcached_delete_by_key (ptr=0x4e2b5490,
master_key=0x7f0f9bd744b0 "1:206384:48", master_key_length=11,
key=0x7f0f9bd744b0 "1:206384:48",
   key_length=11, expiration=0) at memcached_delete.c:71
#10 0x000000000067d970 in memcached_delete (ptr=0x4e2b5490,
key=0x7f0f9bd744b0 "1:206384:48", key_length=11, expiration=0) at
memcached_delete.c:6

(gdb) up 5
#5 0x0000000000681841 in memcached_purge (ptr=0x4e2b5720) at
memcached_purge.c:56
56 result_ptr);
Current language: auto; currently c
(gdb) l memcached_purge.c:35
30 }
31 WATCHPOINT_ASSERT(ptr->fd != -1);
32
33 uint32_t no_msg= memcached_server_response_count(ptr) - 1;
34 if (no_msg > 0)
35 {
36 memcached_result_st result;
37 memcached_result_st *result_ptr;
38 char buffer[SMALL_STRING_LEN];
39
(gdb) l memcached_purge.c:56
51 for (x= 0; x < no_msg; x++)
52 {
53 memcached_result_reset(result_ptr);
54 memcached_return rc= memcached_read_one_response(ptr, buffer,
55 sizeof (buffer),
56 result_ptr);
57 /*
58 * Purge doesn't care for what kind of command results that
is received.
59 * The only kind of errors I care about if is I'm out of sync with the
60 * protocol or have problems reading data from the network..
(gdb) p no_msg
$5 = 4294967295
(gdb) p (uint)-1
$6 = 4294967295

memcached_delete() try purge pending responses but there is nothing
(i.e. memcached_server_response_count == 0).

--- libmemcached-0.30/libmemcached/memcached_purge.c 2009-05-20
22:08:20.000000000 +0400
+++ libmemcached-0.30-my/libmemcached/memcached_purge.c 2009-07-10
19:42:08.000000000 +0400
@@ -30,8 +30,8 @@
  }
  WATCHPOINT_ASSERT(ptr->fd != -1);

- uint32_t no_msg= memcached_server_response_count(ptr) - 1;
- if (no_msg > 0)
+ uint32_t no_msg= memcached_server_response_count(ptr);
+ if (no_msg-- > 1)
  {
    memcached_result_st result;
    memcached_result_st *result_ptr;