Electoral reform maven Rick Hertzberg has a long, informative but inconclusive post about the fight for the popular vote in the Democratic primary. The one criticism I'd make is that he too readily concedes the fairness of adding the Florida and Michigan votes to Hillary Clinton's totals. As we know, neither candidate campaigned in either state, and Obama did not appear on the ballot in Michigan. Hertzberg argues that the fair thing would be to count most of the "uncommitted" vote in Michigan for Obama, but that still doesn't quite balance the scale. The candidate whose supporters have to figure out that "uncommitted" is the best expression of their preference is obviously at an enormous disadvantage.
Now, I'll concede that completely excluding Michigan and Florida isn't perfectly fair, either. Clinton surely would have won Florida under normal conditions. On the other hand, her margin would very likely have dropped, as it has everywhere the two candidates have campaigned. And I strongly doubt Clinton would have won Michigan under normal conditions by the margin she prevailed over "uncommitted," or even at all. Her margin in Michigan exceeded her margin in Ohio. But Michigan (my native state) has a higher black population than Ohio, and a different political culture. Michigan is more of an upper-Northwest state, while Ohio has significant numbers of Southern-Appalachian "Jacksonian" Democrats. If they held a re-vote in Michigan, I'd bet on Obama. (Unlike Florida, where Clinton would be the clear favorite.)
So, if you really want to gauge the overall sentiment of the Democratic electorate, you'd have to count Florida and Michigan but severaly discount Clinton's margin there -- indeed, I'd probably exclude Michigan altogether. Exactly how much to discount them, I couldn't say.
I think the overall takeaway point is that there are lots of problems with tabulating the popular vote in the primary. The fact that caucus states produce far fewer votes than primary states is another severe distorting influence. At best, the popular vote is a very rough and approximate measure.