A multi-sink wireless sensor network (WSN) is composed of a large number of sensors deployed in a wide area and there are some special sensors referred as sinks which play the role of data collector. Sinks are connected to internet so that they can forward data to the data collecting server. Thus each sensor can forward the information they sensed to any one of these sinks. Since sensors located in different positions may bear different communication responsibilities, e.g. forwarding the data generated by some other sensors, the loading or energy consumption of each sensor in network is heterogeneous. Obviously, some sensors may die earlier than the others and thus cause the bottleneck phenomenon in the network. When the bottleneck phenomenon occurs, most of the data from sensors cannot be delivered successfully to the sinks. In this paper, we evaluate the energy consumption as well as the load distribution of sensors deployed uniformly in multi-sink WSNs. With the knowledge of load distribution of a network, we find that not only sensors located near around the sinks but also those in the forwarding paths from the farthest sensors to the sinks will consume more energy. A forwarder selection method is proposed to leverage the loading of sensors in the longest forwarding path. Moreover, a sensor deployment strategy by density control is also proposed to balance the loading of sensors in different tiers, prolong the network lifetime, and increase the successful data delivery as possible. Simulation results show that the data delivery and survival rates are significantly improved by the proposed methods.