requested offset < lowest offset ?

Did anybody see this before?

97/05/26 23:12:19| WARNING: Forwarding loop detected for '
http://mssjus1.www.conxion.com/msdownload/netmeeting/en/Mnm2095.exe'
97/05/26 23:12:19| --> 1.0 proxy.easynet.fr:8080 (Squid/1.1.9), 1.0
proxy06.isracom.co.il:3128 (Squid/1.1.10), 1.0 proxy.isracom.co.il:3128
(Squid/1.1.10)

 The <proxy.easynet.fr> is a sibling of <proxy.isracom.co.il> which
in turn is a sibling of <proxy06.isracom.co.il>(we have 5 squids more).
It works in this combination since the end of November(of course
we upgrade the versions :-). The message from above never appeared,
for those caches, although I've seen it earlier for other combinations.
The disks of <proxy06> were 92% full(~11GB). Today, after the above
message in logs, the <proxy06> has trashed about 2GB of its cache,
within ~4 hours!
 The cache space has gone into the cache.log(I keep the logs within the
same partition as cache's one):

97/05/26 23:12:22| storeStartDeleteBehind: '
http://mssjus1.www.conxion.com/msdownload/netmeeting/en/Mnm2095.exe' at
1056768 bytes

 and after that, ~2GB of what You may see below:

97/05/26 23:12:23| WARNING: storeClientCopy: requested offset < lowest
offset
97/05/26 23:12:23| --> 0 < 4096
97/05/26 23:12:23| -->
'http://mssjus1.www.conxion.com/msdownload/netmeeting/en/Mnm2095.exe'
97/05/26 23:12:23| WARNING: storeClientCopy: requested offset < lowest
offset
97/05/26 23:12:23| --> 0 < 4096
97/05/26 23:12:23| -->
'http://mssjus1.www.conxion.com/msdownload/netmeeting/en/Mnm2095.exe'
 
 ...so on untill ~3:30, when I've catched it.

Nothing "wrong" happened; after I've rerun it, the Squid started as
usually and runs OK, but...
Not a bad performance test - fill up 2GB ,the same time deleting
objects behind(?), within 4hrs, isn't it? :-(
 The system is Linux, kernel v.2.0.27, patched to have 2048 FDs.
Dual PPro-200, 256MB RAM, WSCSI 6x2GB stripped in RAID0, "personal"
_named_, dedicated to Squid, ~40K connections/hour.

-- 
 Gregory.
06