If you read the Cisco best practises it is recommended to use unique Native VLAN's for each Trunk. For example Distribution switch #1 has 10 trunks to access switches, trunk 1 will carry data vlan 10, and voice vlan 110, the Native vlan will be 1010, trunk #2 will carry data vlan #11, voice vlan 111 and Native vlan 1011 etc. The Native VLAN's should NOT be configured with any Layer-3 interfaces, neither should they be allowed on the trunk itself, they do however need to exist at layer-2; i.e.:
vlan 10
vlan 110
vlan 1011
!
interface GigabitEthernet3/1
switchport
switchport trunk encapsulation dot1q
switchport mode trunk
switchport nonegotitate
switchport trunk native vlan 1010
switchport trunk allowed vlan 10,110
!
The security issue is the ability to hop between vlans with specifically crafted packets. Quote from CCO:
"When double-encapsulated 802.1Q packets are injected into the network from a device whose VLAN happens to be the native VLAN of a trunk, the VLAN identification of those packets cannot be preserved from end to end since the 802.1Q trunk would always modify the packets by stripping their outer tag. After the external tag is removed, the internal tag permanently becomes the packet's only VLAN identifier. Therefore, by double-encapsulating packets with two different tags, traffic can be made to hop across VLANs."
This is quite well documented and has been exposed although not explotited very much.
HTH
Andy