I did some quick capacitance measurements using the 34461A's new feature compared to two other devices.
With the more instruments I get, the less know about the parts I'm measuring.
Just for a giggle, I tried repeating your experiment, using a cheap 100nF leaded capacitor.
Averaged over 157 samples, my 34465A quotes a mean value of 106.87nF and std dev of 0.07nF.
My Fluke 89 IV says 103.5nF.
My Fluke 289 says 106.5nF. All so far, so good.
However: my HP 4395A impedance analyser can't make head nor tail of it. The cap's amplitude and phase response above a few MHz is repeatable but complicated, making the equivalent circuit very sensitive to small measurement errors. Each time I press the button which calculates an equivalent circuit for the connected device, I get a different answer, and the answers only really agree on the cap's resonant frequency, which is about 5.42 MHz.
If I limit the analysis to 500kHz, I get C = 97.64nF, ESR = 53.6mOhm, and ESL = 13.1nH.
Not sure what the moral of the story is here, if there even is one. Probably something about measuring components under conditions which reflect how they'll actually be used. Also, a cheap 100nF leaded capacitor makes a crappy decoupler.