@ sg: Please have a look.

I'm now using version 3.1.0 and the problem also exists there.

There seems to be a difference when running the server process on Windows or Solaris. When the server 
disappears on Solaris the client is not locked, while when the same happens with a server process on 
Windows the client remains locked.

I am not sure what the fix should be here, since something goes wrong here with
the server.
I assume that this can be fixed easily by restarting the client?

It is difficult to reproduce. I don't know the code, but there may be a synchronized section too many in 
the client stub. Restarting the client is indeed a workaround.

I've run into this as well, and our load tester seems to be able to reproduce it fairly easily. Will update more details on that once I have them.

For my code we're using the 'client' inside of a server process. This makes restarting the client a bit less desirable.

It seems to happen when attempting to open a new connection or an instance of OpenOffice that has locked up or is otherwise in a bad state. It seems to just sit on a wait hoping to be notified by another thread. It just never happens.

I've updated a copy of writeRequest to look like this:
// @see IProtocol#writeRequest
    public boolean writeRequest(
        String oid, TypeDescription type, String function, ThreadId tid,
        Object[] arguments)
        throws IOException
    {
        if (oid.equals(PROPERTIES_OID)) {
            throw new IllegalArgumentException("illegal OID " + oid);
        }
        synchronized (monitor) {
            while (!initialized && state != STATE_TERMINATED) {
                try {
                    monitor.wait();
                } catch (InterruptedException e) {
                    Thread.currentThread().interrupt();
                    throw new RuntimeException(e.toString());
                }
            }
            if (state == STATE_TERMINATED) {
                throw new DisposedException();
            }
            return writeRequest(false, oid, type, function, tid, arguments);
        }
    }

Adding in the state != TERMINATED. This appears to have fixed the problem so far, but I think it still needs a timeout on the monitor.wait() call so it doesn't sit there indefinitely waiting.

When I detect a stuck instance my code kills the OpenOffice Instance, and also calls close on the bridge object associated with this. That code flips the state to STATE_TERMINATED and notifies on monitor. I'm not sure if that will work for all users, so a timeout would probably be needed here as well.