This paper addresses the problem of load-balancing in a typical mobile edge computing scenario, where mobile users need to select one of the edge servers in their vicinity to offload their tasks. To achieve load balancing, while also limiting the control overhead, we propose delta-DJSQ, a fully-distributed joint-the-shortest-queue policy, where servers provide sporadic status updates over unreliable channels, and users select the least loaded server according to their local belief. We formally prove that the necessary conditions of system stability are also sufficient, namely, the system is stable when the task generation rate is lower than the task service rate, the status update rate is positive, and the probability that a user receives a status update is positive as well. We show that delta-DJSQ is robust to update delays and to message losses, and achieves competitive performance results compared to static load-balancing solutions, both for homogeneous and for heterogeneous servers.