Over the past decade, there has been much focus on mobile ad-hoc sensor networks. The mobility alleviates several issues relating to sensor network coverage and connectivity, whereas aggravates the difficulties of applications such as target tracking. Traditional solutions always localize the sensors first, and then track the target. In contrast, cooperative simultaneous localization and tracking (CoSLAT) adopts both the sensor-target and the inter-sensor observations to simultaneously refine the target and the sensor estimates. We propose a distributed variational filtering (VF) algorithm for CoSLAT, which greatly cuts down the estimate errors, while having nearly the same complexity as the traditional particle filtering (PF) algorithm. In addition, the update and the approximation of the a posteriori distribution are jointly performed by the VF, yielding a natural and adaptive compression. Since the temporal dependence is reduced from a great number of particles to one Gaussian component, the communication cost is significantly diminished.