We have conducted a survey for warm-hot gas , traced by O vi absorption in the spectra of 5 high-redshift quasars ( 2.2 < z < 2.8 ) observed with Keck I/HIRES . We identify 18 O vi systems , 12 of which comprise the principal sample for this work . Of the remaining six systems , two are interpreted as ejecta from the QSO central engine , and four have ionization conditions affected by proximity to the background QSO . Of the 12 intergalactic O vi absorbers , 11 are associated with complex systems showing strong Ly- \alpha ( N _ { { H \mbox { \tiny I } } } > 10 ^ { 15.2 } cm ^ { -2 } ) , C iv , and often other lower ionization species . We do not detect any lines that resemble photoionized , enriched gas associated with the lowest density regions of the Ly- \alpha forest ( 13.5 < \log N _ { { H \mbox { \tiny I } } } < 14.5 ) . Not all of the systems lend themselves to a straightforward determination of ionization conditions , but in general we find that they most closely resemble hot , collisionally ionized gas found near regions of significant overdensity . The extent and gas density of the intergalactic O vi absorbing regions are constrained to be L \leq 200 kpc and \rho / \bar { \rho } \geq 2.5 . This was calculated by comparing the maximum observed O vi linewidth with the broadening expected for clouds of different sizes due to the Hubble flow . For the median observed value of the Doppler parameter b _ { { O \mbox { \tiny VI } } } = 16 km/s , the inferred cloud sizes and densities are L \sim 60 kpc and \rho / \bar { \rho } \sim 10 - 30 . The clouds have at least two distinct gas phases . One gives rise to absorption in photoionized C iv and Si iv , and has temperatures in the range T = 20 , 000 - 40 , 000 K , and overdensities of \rho / \bar { \rho } \geq 100 . The second phase is traced only in O vi absorption . Its temperature is difficult to constrain because of uncertainties in the nonthermal contribution to line broadening . However , the distribution of upper limits on the O vi , C iv , and Si iv temperatures indicates that the O vi thermal structure differs from that of the other ions , and favors higher temperatures where collisional ionization would be significant . The O vi systems are strongly clustered on velocity scales of \Delta v = 100 - 300 km/s , and show weaker clustering out to \Delta v = 750 km/s . The power law slope of the two-point correlation function is similar to that seen from local galaxy and cluster surveys , with a comoving correlation length of \sim 11 h _ { 65 } ^ { -1 } Mpc . The average Oxygen abundance of the O vi systems is constrained to be [ O / H ] \geq - 1.5 at z \sim 2.5 , about 10 times higher than the level observed in the general IGM . Two production mechanisms for the hot gas are considered : shock heating of pre-enriched gas falling onto existing structure , and expulsion of material by supernova-driven galactic winds . Comparison between the observed numbers of O vi systems and expectations from simulations indicates that infall models tend to overproduce O vi lines by a factor of \sim 10 , though this discrepancy might be resolved in larger , higher-resolution calculations . Known galaxy populations such as the Lyman break objects are capable of producing the amount of O vi absorption seen in the survey , provided they drive winds to distances of R \sim 50 kpc .