killing a top-level-listener

I am using #'system::listener-top-level in code that implements a 
remote-repl in a delivered server image.
I have a difficult to debug problem with #'system::listener-top-level.

A standalone test (with only public functions ;-) is as follows:

---------file: deliver.lisp---------

(in-package :cl-user)

;; bring us to the latest official LispWorks patch level

(load-all-patches)

;; make sure we have the comm module

(require "comm")

;; start the actual delivery process

(deliver #'(lambda ()
              (comm:start-up-server :service 10243)
              (loop (sleep (* 60 60))))
          "test-image"
          0
          :multiprocessing t
          :keep-pretty-printer t
          :keep-debug-mode t
          :keep-conditions :all
          :keep-load-function :full)

(quit)

---------eof----------

Now deliver an image like this:

/Applications/Lispworks/lispworks-tty -site-init - -init deliver.lisp

Next start the image ./test-image

And use telnet in another terminal to connect to it:

[sven@voyager:~]$ telnet localhost 10243
Trying ::1...
telnet: connect to address ::1: Connection refused
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
"foo"

CL-USER 1 > (mp:ps)

#<MP:PROCESS Name "10243 server" Priority 3 State "Waiting for 
connection">
#<MP:PROCESS Name "foo" Priority 3 State "Running">
#<MP:PROCESS Name "Initial delivery process" Priority 0 State "Sleeping 
on mailbox">
#<MP:PROCESS Name "The idle process" Priority -536870912 State "Running 
(preempted)">
NIL

CL-USER 2 > foo

Error: The variable FOO is unbound.
   1 (continue) Try evaluating FOO again.
   2 Specify a value to use this time instead of evaluating FOO.
   3 Specify a value to set FOO to.
   4 (abort) Return to level 0.

Type :b for backtrace, :c <option number> to proceed,  or :? for other 
options

CL-USER 1 : 1 > Terminated

After entering a sub listener handling an error, kill the telnet 
process from yet another terminal.
This make the image unustable: cpu goes to 100%, and it is no longer 
possible to connect to it.
Platform is LWM 4.3.7 on Mac OS 10.3.6.
The problem does not occur when the listener is at level 0.

I need a remote-repl style system to access a running server image for 
maintenance tasks,
that is robust (i.e. does not kill the server image when something goes 
wrong).

Sven