O VI absorption is observed in a wide range of astrophysical environments, including the Local ISM, the disk and halo of the Milky Way, high-velocity clouds, the Magellanic clouds, starburst galaxies, the intergalactic medium, damped Lyman-α systems, and gamma-ray-burst host galaxies. Here a new compilation of 775 O VI absorbers drawn from the literature is presented, all observed at high resolution (instrumental FWHM≤20 km s −1 ) and covering the redshift range z=0-3. In galactic environments [log N (H I) 20], the mean O VI column density is shown to be insensitive to metallicity, taking a value log N (O VI)≈14.5 for galaxies covering the range -1.6 [O/H] 0. In intergalactic environments [log N (H I)<17], the mean O VI component column density measured in datasets of similar sensitivity shows only weak evolution between z=0.2 and z=2.3, but IGM O VI components are on average twice as broad at z=0.2 than at z=2.3. The implications of these results on the origin of O VI are discussed. The existence of a characteristic value of log N (O VI) for galactic O VI absorbers, and the lack of evolution in log N (O VI) for intergalactic absorbers, lend support to the "cooling-flow" model of Heckman et al. (2002), in which all O VI absorbers are created in regions of initially-hot shock-heated plasma that are radiatively cooling through coronal temperatures. These regions could take several forms, including conductive, turbulent, or shocked boundary layers between warm (∼10 4 K) clouds and hot (∼10 6 K) plasma, although many such layers would have to be intersected by a typical galaxy-halo sightline to build up the characteristic galactic N (O VI). The alternative, widely-used model of single-phase photoionization for intergalactic O VI is ruled out by kinematic evidence in the majority of IGM O VI components at low and high redshift.