I found this article quite late, (1 year and 3 pages into comments)
The issue as discussed for a long time in the forums, dating back to the release of the K10D , as Steve says relates largely to the focusing screen and the way the meter takes light off the focusing screen.
This was characterized for several cameras with the following chart
What was found is that most cameras with stopped down metering were accurate between F4 and F5.6. There is a reason for that, it is the native aperture range of most consumer “Kit Lenses”.
Note in the chart, 18% grey exposure is about 128 greyscale on the histogram. Each F-Stop is about 40 greyscale.(at least between 30 and 225, then it becomes quite nonlinear.
Yes the metering improved greatly on the K10 with the focusing screen from the *istD but there were also other issues.
The use of teleconverter with feed through aperture pins also had metering issues, due to the reported aperture being different than the real native aperture. I demonstrated this with my sigma 70-200/2.8 with 1.4x and 2x teleconverter.
But that is not the end of exposure woes, my Tamron 28-75/2.8 for example has an error in the aperture movement that starts out under exposing by 1/2 stop wide open, to over exposing by 1/2 stop at F22. Exposure vs stops is a straight line between these two ends of the curve.
The bottom line is you should test your combinations to know how they work.
It takes about 10 minutes per lens / camera combo to do this.
Edit note, one issue with the K10D no one talks about, as it was changed in firmware, is that the initial stop down metering period was too short, and they metered before the lens was stopped down. The length of time for stop down metering was only about 1/2 the time of then*istD at the onset. But later software revert d it back to the same time. I know, because I reported the much shorter metering time to Pentax myself