Can early voting numbers tell us anything?

Last week, I cited some early voting numbers collected by Michael McDonald that suggested a Democratic advantage for the midterm elections.  This post at National Review's Online, however, compares these early voting numbers with those of 2008 and finds the Democrats wanting this year.  That is, even though Democrats still outnumber Republicans among early voters, their advantage has waned significantly:

The average of these states show that early voting has shifted from a D+16.6 partisan split to a D+1.7 partisan split for a Republican gain of +14.9% since 2008.

Over at the Washington Post, however, Karen Tumulty reports that the partisan balance among early voters is almost identical to that of 2006, when Democrats took over both chambers.

So are the early voting numbers a good sign for Democrats or Republicans?  What's the proper basis of comparison -- 2008 or 2006?

Probably the most important question here, though, is just how predictive are early voting statistics of actual election outcomes?  I've not seen a serious study of this (please let me know if you have), but my guess is they're not terribly helpful.  For one thing, widespread early voting is a relatively recent phenomenon, so we don't have a whole lot of data here.  For another, as McDonald reminds us, early voting doesn't tell us how people voted.  All we know is their party affiliation.  Yes, any registered partisan who bothers to vote early is extremely likely to vote her party registration.  But a) this doesn't tell us the percentages of registered partisans who will ultimately vote; and b) this doesn't tell us a thing about the preferences of independents, who comprise somewhere around a fifth of the early electorate thus far.

At this point, only around 5 to 10 percent of those who are casting a vote in 2010 have voted, and I feel confident in saying that that 5 to 10 percent is poorly representative of the ultimate electorate.