[TomB_tx]

I was recently doing a series of tripod-mounted shots to illustrate the change in DOF with aperture setting, using a K series 50 f1.2, taking the same close-up while progressing through f-stops, using manual mode and the green button to set exposure.
Exposures were quite consistent from f1.2 to f5.6, then began to underexpose progressively from f8.0 (slightly) through f22. I think this underexposure with the green button may have been mentioned in posts before, but I hadn't encountered it - probably because I seldom stop down more than f8.
So I've experimented further. Using the same setup, metering first with the green button, and then checking with manual metering stop-down preview, I confirmed that both methods match from f1.2 to f5.6 with this lens, then manual metering with the preview lever shows longer exposure is needed compared to the green button result.
Repeating with an SMC-M 50 1.4 showed that both methods agreed from f1.4 through f8.0, and stopping further caused the green button to underexpose progressively.
Using an SMC-M 50 f1.7 the exposures agreed through a half-stop more.
Using an SMC-M 100 f4 macro, the green button matched manual readings through f16, then underexposing for f22 and f32.
So I'd conclude that the green button, on my K-5 at least, doesn't synchronize the instant of reading carefully enough for the lens to stop down fully, if set for the smaller apertures. If it did, the reading would match that taken using the preview lever.
This only becomes an issue when stopping way-down, which is rarely used, except when you need great depth-of-field, more typically with macro work. When doing so, metering in manual mode is more accurate with the preview lever than with the green button.
Otherwise, since best lens performance is usually wider than f8, the green button works fine.
I'd be interest to hear if others can confirm this on their K-5s. Just meter a K-mount (M series) lens in manual mode with the green button, stopped down to f16, and then pull the preview lever and see if the meter display is centered using the shutter speed the green button selected. You can try different f-stops to see where they match.

[demp10]

How much ambient light did you have when doing the experiment?

The camera theoretically can measure down to 0EV (F1.0, 1 sec at 100 ISO) but as you close a manual lens you are losing EVs while metering. With an auto aperture lens, the camera measures wide open so all the available light reaches the exposure sensor and then calculates the proper exposure for the aperture selected. When you use the green button, the lens is stopped down to the selected aperture before the measurement and the light the reaches the sensor is reduced accordingly. It is possible under moderate or poor lighting condition, to close the lens to a point that the light that reaches the sensor is less that 0EV (or very close to produce inaccurate results).

Have you tried in very bright light (EV 13-15) and see if you get the same results?

[Anvh]

I believe this is in the manual.

The lightmeter is only able to meter the light only so far, so with f/16 the amount of light hitting the meter is simply to low. Like Demp explained.

One trick around it is to use a lower aperture for metering, hit the AE lock button and then move the aperture to the value you want.

[JinDesu]

Theoretically, you could calculate it, no?

Or as Anvh says, AE lock will calculate it for you.

[demp10]

Theoretically, you could calculate it, no?

Sure you can. First measure wide open by hitting the green button. Then close the aperture to the desired f-stop and lower the shutter speed by the same number of stops.

[TomB_tx]

However, it measures accurately using the manual preview lever, so it can't be the light level. Also, the green button is accurate to smaller f-stops with slower lenses than fast lenses. If it were light level related, it would be the same fstop limit.
It is more related to how many stops closed from wide open. Notice as the lens fstop gets slower, it remains accurate to slower stops, but approximately the same ratio to the wide open aperture. That suggests the green button measures too quickly as the lens is stopping down.

[demp10]

I run a quick test on K-7 with the smc Pentax-M 50mm F1.7 and I got some strange results. The measurements were identical using the green button or the preview level but the actual values were not as expected. Actually it overexposed on most settings assuming that the wide open measurement is the correct.

The test was far from a scientific one. When I'll have some time I'll repeat it with more lenses and a more stable setup.

These are the measurements:
F1.7 = 1/640
F2 = 1/320
F2.8 = 1/250
F4 = 1/80
F5.6 = 1/30
F8 = 1/15
F11 = 1/5
F16 = 1/2.5
F22 = 1/2