Combining the data shown in Figure 1 with that shown in Figure 3 allows us to determine that, in the most recent release of the MSRT (March 2006), the rate of infected computers per executions of the tool was 0.28 percent. In other words, the tool removed malware from approximately one in every 355 computers on which it ran. The average rate across all releases from June 2005 to March 2006 is similar, at 0.32 percent, or approximately one in every 311 computers. This infection rate has remained relatively constant across the measurable releases, with the high being 0.4 percent in August 2005 and the low being 0.24 percent in September 2005.
That would make these articles, and the erroneous interpretation of the statistics, FUD.
OK - so the answer is, the number is really really small.
Close enough? :~D