Wi-Fi devices are being increasingly deployed for both traditional high-throughput and new IoT applications. Thus, highly heterogeneous Wi-Fi devices with different configurations and capabilities will be co-located in dense deployments. However, it is not yet fully understood how such devices share the spectrum among each other and what performance they achieve, especially in dense emerging deployments. This paper presents extensive throughput measurements from a dense indoor IEEE 802.11ac testbed with heterogeneous commercial devices. Some links use high-throughput Asus devices with sophisticated MIMO configurations and others use single-antenna Raspberry Pis (RPis), as simpler IoT devices. Moreover, these links are configured to operate on different co- and adjacent channels of different widths. We show that there is a non-trivial interplay between PHY capabilities and contention at the MAC layer, with unintended consequences. Namely, adjacent channel interference (ACI) causes asymmetric channel contention where only Asus devices defer transmissions, whereas RPis do not, likely due to their different receiver filters. Two major consequences are: (i) in heterogeneous deployments the Asus links often have a throughput that is tens of Mbps lower than that of the RPi links, despite the more sophisticated MIMO capabilities of the Asus devices; and (ii) contention due to ACI among devices operating on narrow channels affects contention also with devices operating on wide overlapping channels and determines whether the latter starve or not. Our results suggest overall that better knowledge of the unintentional effects due to PHY-MAC interactions is needed, in order to explicitly take this into account when designing fair and predictable spectrum sharing mechanisms.