Enhanced Inter-Cell Interference Coordination (eICIC) is a key ingredient to boost the performance of co-channel Heterogeneous Networks (HetNets). eICIC encompasses two main techniques: Almost Blank Subframes (ABS), during which the macro cell remains silent to reduce the interference, and biased user association to offload more users to the picocells. However, its application to realistic irregular deployments opens a number of research questions. In this paper, we investigate the operation of eICIC in a realistic deployment based on three-dimensional data from a dense urban European capital area. Rather than the classical semi-static and network-wise configuration, the importance of having highly dynamic and distributed mechanisms that are able to adapt to local environment conditions is revealed. We propose two promising cell association algorithms: one aiming at pure load balancing and an opportunistic approach exploiting the varying cell conditions. Moreover, an autonomous fast distributed muting algorithm is presented, which is simple, robust, and well suited for irregular network deployments. Performance results for realistic network deployments show that the traditional semi-static eICIC configuration leads to modest gains, whereas the set of proposed fast dynamic eICIC algorithms result in capacity gains on the order of 35-120% depending on the local environment characteristics. These attractive gains together with the simplicity of the proposed solutions underline the practical relevance of such schemes.