For what it's worth, I was looking at the piezo output on a scope with a 1 M resistor in parallel with the piezo leads. Of course, the exact output depends on what surface the piezo is mounted on and how hard you knock. I had mine mounted on 1/4 inch plywood. The dominant frequency of vibration appeared to be about 500 Hz. It decayed rapidly over a few cycles. At 500 Hz, one cycle is just 2 milliseconds, so having a delay of 100 milliseconds between each reading of the analog input is way too long.
The comment on the code indicates that the delay is to "avoid overloading the serial port buffer". If that is so, then the delay only need occur when a knock is detected and serial output is sent.
As I worked on my first project using a piezo as a knock sensor, this bug caused me lots of extra work. I was using the sample code to test that the knock sensor was working. It appeared to be very insensitive. So I made several modifications including an op-amp added to the piezo output, and a mass glued to the piezo. In reading the forums, it appeared that others were also having trouble with low or inconsistent knock sensitivity when using the code from this knock tutorial.
Thanks for your attention to this. I'm hoping to help others avoid this same problem.