Well that may explain it... That's the sentence in the datasheet I must have missed!
I'm applying the 1 Mhz to the Fin pin via a 1nf capacitor. The circuit I've constructed is very similar to the one described here:
http://hem.bredband.net/noda/freqmanual.pdf. I'm using a different CPU (a parallax propeller) and I'm driving it from 3.3V. This particular project claims to work down to 5 Mhz. I have tried injecting signals as high as 16 Mhz into mine, and still get the divide-by-8 behavior. I don't have the cp_out line connected to a microcontroller, but rather have it going to my scope and/or frequency counter
I suspect they're not using the chip for it's intended behavior and maybe this is how they're operating it out of the datasheet's stated range -- they're not using the oscillator circuit at all, and are using it only as a prescaler based on the N counter. Code that I found in github suggests they operate the device in test mode (TEST=1, CP_TRI=1, PD_POL=1).
The output when observing with the scope is noisy. I'm wondering if shoddy construction technique in my prototype is to blame, and input noise is being detected and triggering the counter.
Scott