In this work, we revisit classical packet radio networks with a modern treatment of physical-layer (PHY) procedures, medium-access (MAC) and geographic channel-driven routing protocols. Our network model assumes that nodes are randomly distributed on the plane according to a homogeneous spatial Poisson process, using which we provide a novel representation of interference statistics resulting from packet collisions. Using this representation, we develop a cross-layer analysis methodology which allows multi-hop routing protocols to be treated using generic information-theoretic models for the underlying PHY/MAC procedures. These models inherently characterize modern procedures such as channel code rate adaptation, incremental redundancy and packet combining/capture. These models further allow for the assessment of the tradeoff between spatial throughput, measured in bit-meters per signal-space dimension, the range of each transmission and the average transmission delay. A generic formulation based on system parameters, such as system bandwidth, propagation models, etc., is given to analyze this tradeoff in an operational setting which can be used to build system simulators for such networks. Finally, from a purely PHY perspective, the results of this work show that coding and incremental retransmission provide a means for reliable communication coupled with a completely decentralized multiple-access strategy.