Interestingly, the metering of the camera reflects the nominal difference (factor 8) the filter should make.
How reproducible is the effect on the pictures when you take several pictures in a row? Does it also happen when you open the aperture all the way? I've seen lenses where a bad aperture mechanism resulted in random exposure variations...
Edit: If the EXIF tags are to be believed, the first picture was taken at 1/125s, not 1/100s, which would make only 2.7 stops difference. Also, as grantyale points out in the post below, make sure you don't use evaluative metering - it is pretty much a random number generator.
To determine the effect of the filter, you could try a series of different exposures in manual mode. That should make it clearer what part of your troubles can be attributed to the filter, and what part to the camera/lens.