DNSCACHE / High CPU LOAD & Socket error ?

Hello,
I'm using BIND for DNS CACHE.

Using debian package, version is: BIND 9.2.1

Sometimes, CPU incread to 100% and we don't know why...

When trying to stop named properly , it didn"t... only a kill worked.

Because we have lot of queries (~ 500 queries /s) , we tuned the named conf with this options:

recursive-clients 10000;
tcp-clients 5000;

(otherwise, there are "maximum reached" messages).

Other named options are standards:

coresize default;
datasize default;
files default;
stacksize default;

recursion yes;
fetch-glue yes;
fake-iquery no;
notify no;
auth-nxdomain yes;
multiple-cnames yes;
allow-transfer { none; };

cleaning-interval 30;
heartbeat-interval 0;
interface-interval 0;

lame-ttl 600;
max-ncache-ttl 1080;


Here the log during the high CPU Load:

Sep 16 14:44:23 nscache0 named[24351]: socket.c:1100: unexpected error:
Sep 16 14:44:23 nscache0 named[24351]: internal_send: 62.23.125.234#0: Invalid argument
Sep 16 14:44:23 nscache0 named[24351]: client 62.23.125.234#0: error sending response: invalid file

Sep 16 14:44:23 nscache0 named[24351]: socket.c:1100: unexpected error:
Sep 16 14:44:23 nscache0 named[24351]: internal_send: 62.23.125.234#0: Invalid argument
Sep 16 14:44:23 nscache0 named[24351]: client 62.23.125.234#0: error sending response: invalid file
[...]

This message appears 2000 times in log... what does this message mean ?

I saw the same error on Google: http://www.cctec.com/maillists/nanog.../msg00523.html
but don"t give any solution.

The CPU of the box is a PIII 1Ghz with 2Go RAM.

Thanks in advance.

Regards,

Nicolas