We study the coverage performance of multiantenna (MIMO) communications in heterogenous networks (HetNets). Our main focus is on open-loop and multi-stream MIMO zero-forcing beamforming (ZFBF) at the receiver. Network coverage is evaluated adopting tools from stochastic geometry. Besides fixed-rate transmission (FRT), we also consider adaptive-rate transmission (ART) while its coverage performance, despite its high relevance, has so far been overlooked. On the other hand, while the focus of the existing literature has solely been on the evaluation of coverage probability per stream, we target coverage probability per communication link — comprising multiple streams — which is shown to be a more conclusive performance metric in multi-stream MIMO systems. This, however, renders various analytical complexities rooted in statistical dependency among streams in each link. Using a rigorous analysis, we provide closed-form bounds on the coverage performance for FRT and ART. These bounds explicitly capture impacts of various system parameters including densities of BSs, SIR thresholds, and multiplexing gains. Our analytical results are further shown to cover popular closed-loop MIMO systems, such as eigen-beamforming and space-division multiple access (SDMA). The accuracy of our analysis is confirmed by extensive simulations. The findings in this paper shed light on several important aspects of dense MIMO HetNets: (i) increasing the multiplexing gains yields lower coverage performance; (ii) densifying network by installing an excessive number of lowpower femto BSs allows the growth of the multiplexing gain of high-power, low-density macro BSs without compromising the coverage performance; and (iii) for dense HetNets, the coverage probability does not increase with the increase of deployment densities.