As was pointed out, a few days ago on MSNBC Pat Buchanan challenged Richard Cohen on the latter's assertion that it's understandable for blacks to vote for Obama because he's black but "rancid" when whites back Hillary for the same reason. Cohen had preceded Buchanan's challenge by pointing out that Obama was "stumbling to the finish line" without enjoying "a triumphant win" (view the full segment).
Cohen got that one right. Indeed, from the beginning of March onward, Obama's 'momentum' dissipated. As Hillary picked up steam, the previously ubiquitous word all but vanished from media coverage of the campaign. From March through the end of the primaries on June 3, Hillary beat Obama 6,974,327-6,369,076 in total votes*. That's a comfortable 52.3%-47.7% margin.
Despite this post's title and the paragraph above, demography told the story. Braying punditry, campaign speeches, media coverage, and polling data couldn't compete with the value of demographic data in predicting outcomes, especially as the campaign wore on.
I'm currently working on estimates of vote totals by race (non-Hispanic white, black, and Hispanic) and gender for the Democratic campaign in its entirety. Because of the number of caucuses without exit polling data, it's a tedious process.
* I estimated the total turnout for Guam by presuming the population voted in the Democratic caucus at the same rate (9.7%) as those in Puerto Rico voted in the primary there. As caucuses tend to draw smaller numbers than primaries do, this probably overestimates the actual turnout somewhat. Even so the total votes via this estimate only come to 21,105. Obama won by the narrowest margins in Guam (50.1%-49.9%), so if this is artificially boosting his advantage, it's probably doing so by the tens!