This happends randomly while zabbix is running:

 20674:20071109:233103 Query::insert into history (clock,itemid,value) values (1194611462,20580,1189945202891463.750000)
 20674:20071109:233103 Query failed:PGRES_FATAL_ERROR:ERROR:  numeric field overflow
DETAIL:  The absolute value is greater than or equal to 10^15 for field with precision 16, scale 4.

 20674:20071109:233103 Query::select num,value_min,value_avg,value_max from trends where itemid=20580 and clock=1194609600
 20674:20071109:233103 Query failed:PGRES_FATAL_ERROR:ERROR:  current transaction is aborted, commands ignored until end of transaction block

 20652:20071109:233103 One child process died. Exiting ...
 20652:20071109:233105 ZABBIX Server stopped

the SQL query to get the faulty returned:
itemid | hostid | description | lastvalue | lastclock | error
----------------------------------------------------------------------+------
20580 | 10039 | Incoming traffic on interface $1 | 1248.823529 | 1194611446 |
20980 | 10050 | Incoming traffic on interface $1 | 464.357143 | 1194611447 |
20810 | 10048 | Incoming traffic on interface $1 | 12051025.250000 | 1194611462 |

it happened before on other items but all of them point to an incoming traffic (from the Linux template).
the client are using agents of the same version (compiled at the same time) and are all centos4 x86_64 arch (if this helps).

I don't know if it's the agent return some crazy value or the delta processing made on the server which generated this history value but I suppose the server should
not die on insane values....

Problem is because PGSQL refuse any following command after this insert failed, the server decides to die (I think).