According to the gnrc_pktbuf_stats() output the packet buffer contains some chunk when it should be empty. This behavior can be observed when flood-pinging on native to a multicast address (e.g. ping6 100000 ff02::1 0 0) if there is more than one receiver (e.g. another native node connected over a bridge with an IPv6 address configured -> Linux and native replies to the ping). It remains unclear if this problem exist only on native (or only with the Ethernet netdev2 driver) and if it happens only for multicast pings. I was not able to provoke a full packet buffer, neither on native nor on real hardware using the unmodified gnrc_networking example.