Packet-level diversity, or distributing packet transmissions over multiple, diverse channels, offers benefits in improving communication performance and robustness to channel variations. Previous works have analyzed and quantified those benefits, and developed transmission policies to realize them. However, translating those benefits into practice still faces numerous challenges from uncertainty in the adequacy of the channel models used to develop policies, to implementation dificulties in realizing the precise transmission schedules they mandate. This work is a first step in assessing what remains of those benefits once confronted with such practical challenges. Our investigation is carried out over an 802.11 testbed, where diversity is realized through the different frequency bands available for transmissions between hosts and access points. We use the testbed to evaluate the impact of transmission policies, channel characteristics, channel correlation, and various end-system constraints that affect our ability to precisely control transmissions timing. Our investigation reveals that in spite of the many gaps that exist between theory and practice, packet-level diversity still provides a simple solution to improve transmission performance and robustness across a broad range of configurations.