Jeff Stonecash has an interesting paper at <http://www.apsanet.org/ imgtest/PSJuly06Stonecash.pdf>. He shows sharp divisions within income classes on Iraq, He's at Syracuse and easy to find so I'm sure he could give you more info. He also offers evidence countering the standard line that the differences between the two parties have narrowed; by his measures, they've never been so wide.
And there's this:
<http://www.allacademic.com/meta/p17187_index.html>
Diefenbach, D. and West, M. Support for the War in Iraq: American Casualties as a Survey Item Paper presented at the annual meeting of the American Association For Public Opinion Association, Fontainebleau Resort, Miami Beach, FL <Not Available>. 2006-10-05 from http://www.allacademic.com/meta/p17187_index.html
Publication Type: Paper/Poster Proposal Abstract: Support for the war in Iraq has proven difficult to quantify, with respondents often reporting substantial ambivalence if not confusion concerning their opinions re the further prosecution of the war. In an effort to understand the characteristics of those who support further prosecution of the war, and to understand better the nature of their support, a regional random-digit dialed telephone survey applying a willingness-to-pay question to Iraqi war costs was conducted in the first two weeks of October in the Western counties of North Carolina. The survey used a next-birthday method for within- household respondent selection. A total of 493 complete interviews were conducted, with an AAPOR cooperation rate 1 of .48 and a contact rate 2 of .43. Seventy-eight percent of respondents said that they would support no more deaths of U.S. soldiers in Iraq, even after the explanatory introduction to the survey question. Fifteen percent of respondents said they would support any number of deaths in order to fulfill the U.S. mission in Iraq, while 7 % said they would support some intermediate number of deaths, with a modal value of about 100 deaths. A nonlinear regression model was constructed using respondent characteristics to predict the number of U.S. deaths the respondents would find tolerable. Age, income, education, gender, race, newspaper readership, television viewership, and party affiliation were all found to be significant predictors of the number of deaths the respondent would tolerate; the model R2 was .21, and the model was significant at the .01 level. Respondents who were older, watched less television, who were male, had higher income and educational levels, and were white were significantly more likely to support higher levels of U.S. casualties than were other respondents.