OFDMA technology can significantly improve the transmission reliability and efficiency because of its inherent frequency diversity and frequency multiplexing. In our recent work [B.Bai,W.Chen, Z.Cao and K. B. Letaief (2009) ], we have derived the optimal diversity-multiplexing tradeoff for OFDMA systems under the assumption that each subcarrier occupies the entire coherence bandwidth. However in practical OFDMA systems, such as IEEE 802.16, there are many subcarriers in one coherence bandwidth, i.e., each coherence bandwidth is split into multiple subcarriers which brings the correlation of channel gains among these subcarriers. In this paper, we focus on the diversity-multiplexing tradeoff in this kind of OFDMA systems. First, a correlated random bipartite graph is adopted to formulate this problem. To resolve the user conflicts in subcarrier allocation, the maximum proper /-matching method is introduced to minimize the user outage probability with fairness assurance at given multiplexing gains. Based on this model, the optimal diversity-multiplexing tradeoff curve is obtained. Two extreme points are considered: (1) the full diversity gain is the number of coherence bands, i.e., the same as that in point-to-point OFDM systems; and (2) given a coherence bandwidth, the maximum multiplexing gain is equal to the frequency band equally allocated to each user. The random vertices rotation and extension based Hopcroft-Karp algorithm is then proposed as an optimal subcarrier allocation scheme, which can achieve the optimal tradeoff curve with the time complexity of O(S^2.5), where S is the total number of subcarriers.