The problem is that the Chief Justice race was non-partisan on the ballot, so it's not surprising that Connally could do better than Kerry in some Republican areas. If anything, Connally is a Republican name.
Here are the least-squares regression lines at the
precinct level, using Connally share to "predict" Kerry share, for each technology used in Ohio:
(I've suppressed the individual points because they make the plot unreadable.) Ordinary least squares is probably not the best way to model this relationship, but it's a relatively familiar one.
It appears that Kerry may have done "surprisingly well" relative to Connally in DRE precincts (at least the Democratic-leaning ones), versus other precincts, but there isn't much there there. This model is too simple, but not nearly as too-simple as the one that says that Kerry "should have" outdrawn Connally.
Here's another approach: I figured (Kerry votes - Connally votes)/(total votes). That will turn out a bit differently than a model that compares Kerry % in the presidential race to Connally votes in the chief justice race, since fewer people voted in the latter race. The means are:
punch card 0.1163, std error 0.0013, n = 8345
DRE 0.1732, std error 0.0027, n = 1541
opscan pct 0.0271, std error 0.0076, n = 139
opscan ctr 0.0988, std error 0.0029, n = 1341
So again, Kerry seems to do surprisingly well (relative to Connally) in DRE precincts. Over half of those precincts are in Franklin County, and I wouldn't be surprised if the drop-off from presidential to chief justice races tends to run higher in urban areas, but I will leave that for anyone who is curious. Anyway, I don't see a killer fraud argument here, certainly not one that hinges on technology. Your mileage may vary.
If you're a quant, you should read Mebane and Herron's Section VI of the DNC report,
http://macht.arts.cornell.edu/wrm1/Ohio2004/OhioDNC/ -- other parts depending on your interests. And check in with "Time for change."