A large sample of 506 z~6 galaxies (i-dropouts) has been obtained from all the deep, wide-area HST ACS fields: HUDF, GOODS, and UDF-Parallel fields (UDF-Ps). The contamination levels for our selection are very small, <=8% (i.e., >=92% are at z~6). This is the first comprehensive quantitative analysis of such a large sample at z~6 and is used to establish optimal measures of both the luminosity function (LF) and the luminosity density at z~6, and their evolution relative to z~3. Completeness, flux, and contamination corrections are performed in a very systematic way for all three data sets. After performing these corrections, we combine the data to derive a rest-frame continuum UV (~1350A) LF at z~6. We find strong evidence for evolution of the LF between z~6 and z~3. The simplest way to accommodate this evolution is through a brightening of M* (0.7+/-0.4 mag), though less brightening is required if the faint-end slope alpha changes to -1.9. Scenarios, such as density evolution, which do not include this evolution in M* or alpha are ruled out at 99.9999% confidence, demonstrating quite significantly that galaxies at z~6 are lower in luminosity than galaxies at z~3. The substantial evolution in the LF is not accompanied by a large change in the luminosity density. The luminosity density at z~6 (to 0.04 L*_{z=3}) is 0.68+/-0.08x that at z~3. Changes in the mean UV color of galaxies from z~6 to z~3 suggest an evolution in dust content, indicating that the true evolution is substantially larger: the total star formation rate at z~6 is just ~30% of the z~3 value. Despite large uncertainties, our best-fit UV luminosity function is consistent with z~6 galaxies providing the necessary UV flux to reionize the universe. (abridged)