Port 992 (i.e., Secure TN5250), anybody?

Ladies and Gentlemen:

Anybody ever run into a problem with an inbond TN5250
data stream 
(specifically, a response to a Read MDT Fields) going into the
Secure Telnet 
port (992) from a Java-based TN5250 emulator getting cut
off?

It only seems to happen when I'm communicating with a particular host,
doing a 
WRKUSRJOB on a profile with more than one page of jobs, entering an
8 for all 
subfile records, then hitting an AID key. If I leave even one
record 
unselected, or if any of the 8's are already present (thus changing
the length 
of the response data stream), it works fine. If I use a different
host (but 
still on the secured port, with the emulator going through an SSL
socket), it 
works fine. But this particular host, if I enter all 8's (or,
perhaps, 
all-anything-else, but I haven't yet tried it), all fresh, and hit
PgDn, the 
emulator remains in "input inhibited" state until I send a blank
SysRq (even 
weirder: the SysRq gets ignored!), at which point the session
becomes 
"unstuck," and works normally (until it gets another "all-8's"
response to 
that screen).

If I monitor the TCP byte counts on the port in
question from another emulation 
session, I find that the "bytes in"
increments by 1430 bytes (on the other 
host, it increments by 1690 bytes),
but the "bytes out" remains exactly as it 
was before I pressed the AID
key.

And YES, the emulator explicitly flushes its output
stream.

--
J.Lampert