For coordinated multipoint transmission in heterogeneous ultra dense networks (HUDN), clustering and multicell resource scheduling should be carried out jointly. Based on the α-fairness function, the joint optimization problem is formulated that is NP-hard. A two-step joint clustering and scheduling (TS-JCS) scheme is proposed. Considering the extremely imbalanced cell loads in HUDN, the impact of cell loads on resource availability must be concerned, and a novel load aware clustering (LAC) is designed at the first step and solved by game theory. Based on the clustering results, a fair graph-coloring based inter-cell resource scheduling can be employed at the second step to maximize the resource utilization. Simulations are carried out to evaluate the performance of TS-JCS in a three-layer HUDN. It is shown that the utility of TS-JCS is close to the optimal value, demonstrating the high effectiveness of TS-JCS to solve the joint optimization problem. Compared with non-joint schemes, when fairness is considered in TS-JCS with α = 1 and α = 2, the average throughput can be improved by about 30% while keeping similar performance in cell-edge throughput. Finally, it is also shown that in HUDN with imbalanced load, the performance gain of TS-JCS with α = 1 and α = 2 over the baseline scheme is obvious compared to that in regular hexagonal networks with uniformly distributed loads, demonstrating the importance of the load aware scheme LAC for HUDN.