A task loss optimization and load balancing scheduling algorithm under a queuing model was proposed to address the problems of inefficient task scheduling and load imbalance in cloud computing environments. A multi-server environment was abstracted into multiple infinite-capacity GI/M/1/∞ queue models, where task inter-arrival times follow a general distribution and service times adhere to an exponential distribution. By constructing an infinite-state Quasi-Birth-and-Death (QBD) process and applying a two-dimensional Markov process with a matrix geometric solution, steady-state probabilities were derived and performance metrics were quantified. Secondly, a multi-strategy integrated load balancing algorithm was designed: the concept of virtual waiting time was proposed, based on which the loss rate of the quantified task waiting tolerance threshold was calculated to improve the accuracy of task allocation. Combined with the idle queue (join-idle-queue, JIQ) strategy, tasks were prioritized to be assigned to idle servers to reduce waiting time. Through the heartbeat mechanism, the server status was sensed in real time, and parameters were dynamically adjusted to improve resource utilization. Finally, numerical experiments were conducted to compare the performance of the proposed algorithm with that of traditional methods in terms of task loss rate, waiting time, and system stability. The results show that the proposed algorithm significantly improves resource allocation efficiency, reduces the task loss rate, and enhances system stability, providing both theoretical and practical support for load balancing scheduling in high-traffic network environments.