Threshold channel quality index (CQI) feedback helps to achieve a high throughput at a low cost of overhead in multi-carrier systems that have saturated downlink traffic. However, when system is not saturated, the performance will degrade tremendously. In this work we build mathematical model to analyze the phenomenon. We then design traffic-aware threshold CQI feedback schemes to improve feedback efficiency. To further enhance system performance such as fairness, the product of data rate and queue length is applied to determine CQI feedback threshold and BS scheduling. Extensive simulation proves that the proposed scheme leads to a improved throughput and decent fairness in non-saturated systems.