To humanize this statement a little more, let me put this another way: How many innocent people are we willing to kill for the satistfaction of executing the worst criminals? No matter how far forensic science advances, innocent people will die if we have a death penalty.
But we can estimate. The error execution rate has to be at least 1 in 1,000--the "1,000" being Kenneth Boyd and the "1" being Ruben Cantu, who the Houston Chronicle seems to prove that he died for a crime he did not commit. The Death Penalty Information Center lists another eight people as "executed but possibly innocent." That pushes it to about 1 in 100. Estimates for the number of people on death row who have been exonerated range from 25-30 from a prosecutor's estimates to 73 from a University of Michigan study. The maximum possible error rate, depending on very loose assumptions, then surges up to 1 in 30 to 1 in 12. These rates are undoubtedly too high, but they help to establish an upper bound.