We provide an information theoretic perspective on the problem of throughput maximization in a block flat fading wireless data system with codeword lengths restricted to be less than the fade block duration. We assume no channel state information at the transmitter (CSIT) and perfect channel state information at the receiver (CSIR). We explore the tradeoffs between using a single codebook vs. multiple codebooks (rate-splitting) on single input single output (SISO) channels, and scalar coding vs. vector coding for diagonal multiple input multiple output (MIMO) channels. For all log-concave scalar channel fade distributions, we show that using multiple codebooks increases the average throughput of the system when the multiple codewords are transmitted simultaneously in time, frequency and space over the same channel. Splitting the channel orthogonally in time, frequency, or among the inputs of a MIMO system and then transmitting different codewords on each orthogonal sub-channel significantly reduces the achievable average throughput.