A high-performance queueing method to implement a shared queue for collaborative clusters of servers, where each cluster of servers maintains a local queue, and those queues are networked to form a unified (or shared) queue for all those servers in clusters. Requests in an overcrowded queue are forwarded to some other queues, with a randomized algorithm and a message-exchange algorithm between the networked queues. Requests may be forwarded between clusters multiple times. A request is to be accepted with a higher probability by a cluster if the cluster is visited earlier in the order and if the cluster itself has a higher probability to accept the received request. This technique ensures that the requests are forwarded to a queue that has a shorter length (thus a shorter waiting time for first-in-first-out queues) and is closer to the original cluster with a higher probability.