It turns out the actual error is generally outside the commonly states Margin of Error.  The average absolute error, or error regardless of direction, is 7 points compared to the actual Caucus results.

Candte    DMR    Actual    Error    Pcnt Error
Al Gore    56    63    7    11%
Bradley    28    36    8    22%
Bush    43    41    2    5%
Forbes    20    30    10    33%
Keyes    8    14    6    43%
Kerry    26    38    12    32%
Edwards    23    32    9    28%
Dean    20    18    2    11%
Gephrdt    18    11    7    64%
Average            7    28%
Median            7    28%

Statistically the Register's best result was the Gore/Bradley contest, which isn't surprising given that predicting a two candidate contest in caucus is much less complicated.

Nor is there a pattern in inflating or deflating the DMR poll front runner, Gore's result was less understated than Bradley's, George Bush actually underperformed his polling and Kerry over performed.  

The bottom line is this is one data point, not a magic crystal ball. So, for the candidates, staffs, volunteers and reporters, back to work.