I am aware of all the background information to native VLANs, just to repeat here that they are used for 'control traffic', that is CDP / VTP / PAgP / DTP.
On a router, sub-interfaces can be defined to allow different VLAN IDs over a trunk from a connected switch to communicate, each VLAN ID will be on different subnet, the sub-interfaces act as default gateways and have nothing to do with trunking.
Often in prominent study texts and books and answers on forums, it is always stated that you must define the native VLAN on one of the sub-interfaces (or the physical interface in earlier scenarios).
My question is specifically why do we need the native VLAN defined on one of these sub interfaces when the connected VLANs do not need to pass control traffic (e.g. layer 2) to each other?
In every lab I have setup (PT and GNS3), I never have a problem missing out the native VLAN on a sub interface:
router(config-subif) #**encapsulation dot1q 1 native**