In this paper, we investigate how highly dense deployments of small cells perform when full-duplex nodes communicate under composite fading channels. With regard to an arbitrary user of interest, we first identify the impact of the extra interference from co-channel users and how that compares to the intrinsic self-interference. The performance of full-duplex networks is compared to a traditional half-duplex benchmark scenario where the only source of down-link inference is originated by simultaneous transmissions of surrounding base stations. A mathematical framework based on stochastic geometry is used to carry out our investigations. Here, the system performance is assessed by considering various network configurations and distinct fading channel conditions, namely log-normal shadowing and Nakagami-m fading. Using the aforesaid framework, we approximate the aggregate interference at the user of interest and then derive closed-form expressions for the corresponding signal to interference ratio and outage probability. Results show that users in full-duplex mode achieve higher rates even though they are subject to higher interference. It is also shown that the extent of these gains depends on the self-interference cancellation value that can be achieved and how that compares to the inference from other users.