A virtual network is defined as a logical network, in which users exhibit their access behaviors. Virtual networks rely on physical computer networks like the Internet, but have different topologies, and cause significant influence on the physical networks. In this paper, a novel two-tier model is proposed to study influences of virtual networks to Internet collective behavior. It is shown that the queue lengths of the node data packets present phase transition characteristics. Moreover, the phase transition critical point moves to the left and network performance is deteriorated. In a free flow, the nodes are independent of each other or short-range correlative. In the critical state, the nodes are long-range correlative, and there exists a higher power exponent H which means stronger long-range correlation. When the system state is on the right of critical point, virtual network behaviors make the network to present consistent long-range correlative characteristic.