Problems reading more than nine registers at a time

I've got a number of Modbus devices in a laboratory
setup. Each of them is connected to a dedicated Linux
machine (i.e. they don't share the bus). I'm able to
read and write registers and coils on each of them
using jamod - so far so good.

There is, however, one problem: I can only access up to
nine 16-bit registers at a time. If I try to, for
example, read ten registers in a row, the read will
fail because the devices won't answer. This behaviour
is consistent across

- different devices from different manufacturers (Deif
Multiline AGC diesel genset controller, Modicon TSX57
PLC, Deif MIC Switchboard instruments)
- different physical layers (RS232, RS485)
- different baud rates (9600-38400)
- different communication libs (Sun Comm API, RXTX)
- different versions of jamod (1.2rc1, cvs HEAD)

----------------------------------------------
Here's the debug output from a successful read of nine
registers. PLCAITest is essentially SerialAITest with
my own port settings and a few print statements.

olge@syslab-07 ~/devel/Panel-cvs $ java
-Dnet.wimpi.modbus.debug=true -Djava.library.path=.
-classpath comm.jar:.:jamod/src
net.wimpi.modbus.cmd.PLCAITest /dev/ttyS1 1 $((0x130)) 9
net.wimpi.modbus.debug set to: true
Encoding [rtu]
Request: 01 03 01 30 00 09
Sent: 01 03 01 30 00 09 84 3f
Serial port break detected
Response: 01 03 12 13 85 09 30 09 36 09 3a 09 35 0f f3
0f f4 0f f3 0f f3 3e 8a
Response: 01 03 12 13 85 09 30 09 36 09 3a 09 35 0f f3
0f f4 0f f3 0f f3
Word 0=4997
Word 1=2352
Word 2=2358
Word 3=2362
Word 4=2357
Word 5=4083
Word 6=4084
Word 7=4083
Word 8=4083
--------------------------------------------------
...and a failed read of ten registers:

olge@syslab-07 ~/devel/Panel-cvs $ java
-Dnet.wimpi.modbus.debug=true -Djava.library.path=.
-classpath comm.jar:.:jamod/src
net.wimpi.modbus.cmd.PLCAITest /dev/ttyS1 1 $((0x130)) 10
net.wimpi.modbus.debug set to: true
Encoding [rtu]
Request: 01 03 01 30 00 0a
Sent: 01 03 01 30 00 0a c4 3e
Serial port break detected
Last request: 01 03 01 30 00 0a c4 3e
Error reading response
execute try 1 error: I/O exception - failed to read
Sent: 01 03 01 30 00 0a c4 3e
Last request: 01 03 01 30 00 0a c4 3e
Error reading response
execute try 2 error: I/O exception - failed to read
Sent: 01 03 01 30 00 0a c4 3e
Last request: 01 03 01 30 00 0a c4 3e
Error reading response
net.wimpi.modbus.ModbusIOException: I/O exception -
failed to read
at
net.wimpi.modbus.io.ModbusRTUTransport.readResponse(ModbusRTUTransport.java:148)
at
net.wimpi.modbus.io.ModbusSerialTransaction.execute(ModbusSerialTransaction.java:151)
at
net.wimpi.modbus.cmd.PLCAITest.main(PLCAITest.java:124)

-------------------------------------------------

The reply length for a ten register read would be 25
bytes - only ten percent of the length permitted by the
modbus specs.
My suspicion is that the message sent by the master is
being treated as malformed by the slaves. Has anyone
seen similar behaviour, or do you have an idea what the
problem could be? I'd like to get as much bandwidth as
possible out of the serial connections, and splitting
all operations up into nine-register chunks will
produce serious overhead (some of the devices are quite
"rich" in registers).

Thanks
Oliver