In vehicular ad hoc networks, participating vehicles organize themselves in order to support lots of emerging applications. While network infrastructure can be dimensioned correctly in order to provide quality of service support to both vehicle-to-vehicle and vehicle-to-infrastructure communications, there are still many issues to achieve the same performance using only ad hoc vehicle-to-vehicle communications. This paper investigates the performance of such communications for complete applications including their specific packet size, packet acknowledgement mechanisms and quality of service requirements. The simulation experiments are performed using Riverbed (OPNET) Modeler on a network topology made of 50 nodes equipped with IEEE 802.11p technology and following realistic trajectories in the streets of Paris at authorized speeds. The results show that almost all application types are well supported, provided that the source and the destination have a direct link. Particularly, it is pointed out that introducing supplementary hops in a communication has more effects on end-to-end delay and loss rate rather than mobility of the nodes. The study also shows that ad hoc reactive routing protocols degrade performance by increasing the delays while proactive ones introduce the same counter performance by increasing the network load with routing traffic. Whatever the routing protocol adopted, the best performance is obtained only while small groups of nodes communicate using at most two-hop routes.