From the Slate article

They then took the results of their random sample and extrapolated them to the entire country, assuming that their 33 clusters were perfectly representative of all Iraq.

This is incorrect. They left out the area that had the largest number of reported deaths. If we include that area, the number more than doubles.

And they only reported on people the families knew for certain were dead. Their study didn't include the thousands of people who have disappeared.

Bush acknoweledges at least 30,000 civilian deaths based on the Iraq Body Count study, and since the Iraq Body Count authors say that thousands of deaths are not included in their study, we know that the number is thousands more than 30,000.

Add all of that together, and it is not at all a stretch for the authors of the Lancet study to feel confident that their 100,000 number is "conservative".