I've noticed this since the efl-net rewrite, and it's gotten to be a real problem. On my desktop, I run a gadget that uses ecore-con to periodically make client connections and read some data. At some point, it became an issue where I would start getting errors that no more fds could be created because ecore-con was never closing them. Valgrind confirms that a huge number of fds are being leaked, all created from efl_net_socket4().
do you know if the objects get deleted?
I'm pretty sure the objects have "close on destructor" set, thus they should be deleted. If they are not, either some bug with "close on destructor" or some bug in reference counting which let the object dangling...