A fundamental problem arising in dense wireless networks is the high co-channel interference. Interference alignment (IA) was recently proposed as an effective way to combat interference in wireless networks. The concept of IA, though, is originated by the capacity study of interference channels and as such, its performance is mainly gauged under ideal assumptions, such as instantaneous and perfect channel state information (CSI) at all nodes, and homogeneous signal-to-noise ratio (SNR) users, i.e., each user has the same average SNR. Consequently, the performance of IA under realistic conditions has not been completely investigated yet. In this paper, we aim at filling this gap by providing a performance assessment of spatial IA in practical systems. Specifically, we derive a closed-form expression for the IA average sum-rate when CSI is acquired through training and users have heterogeneous SNR. A main insight from our analysis is that IA can indeed provide significant spectral efficiency gains over traditional approaches in a wide range of dense network scenarios. To demonstrate this, we consider the examples of linear, grid and random network topologies.