From IRC:

<zeenix> dcbw: unrefing NMClient calls device-removed for all devices?
<dcbw> zeenix: I'll assume you mean the last ref?
<zeenix> dcbw: yeah and i just checked the code to find out the answer: yes
<zeenix> dcbw: oh i misread the code, it seems device-removed signal is not called :(
<zeenix> dcbw: i think it should
<zeenix> in my device-removed handler, i reset signals handler ids and since this signal is not called on client discpose, i have to do the same on discposing client
<dcbw> zeenix: free_devices(client, FALSE); should be ensuring the signal doesn't got out I think... though if you have references on the devices, they won't go away?
<dcbw> zeenix: so is the issue that when you're cleaning stuff up, it's more code then to go through and clean up every device instead of listening to the signal (which you're already listening to)?
<dcbw> hmm, looks liek a general decision in libnm-glib not to call 'removed' signals unless the object was actually removed by the daemon
<dcbw> zeenix: would you mind filing a bugzilla.gnome.org bug saying this?  we could fix it when we do a libnm-glib2 to kill dbus-glib or perhaps with a "cleanup" signal for child objects