in this paper we propose and investigate a cross-layer multiuser scheduling strategy for the support of heterogeneous traffic in the downlink of a MIMO-OFDMA system. It jointly considers different objectives: maximize the sum-rate on the radio channel, ensure a fair allocation of resources among users belonging to the same traffic class, consider the dynamics of traffic sources by looking at the delay of data packets in the queues, contribute to maximize quality of service figures at the application level. To exploit temporal diversity and to reduce complexity, the ergodic weighted sum-rate is maximized and dual optimization with stochastic approximation is applied to derive on-line algorithms. The numerical results show the capability of the scheduler to allocate physical layer resources according to rate constraints imposed for each different traffic class and with fairness inside each class, even in presence of different channels conditions and different network loads.