PT2399 - how much to trust the datasheet?

[earthtonesaudio]

In the PT2399 datasheet there is a table of delay times, VCO frequencies, etc.  I was thinking the VCO frequency must be a linear function of the delay time, but when I graph the table data I get something rather non-linear.

Now, I would like to use the VCO as an error signal to achieve an accurate tap-tempo device with minimal computation overhead.  But to do that I need an accurate idea of how many VCO cycles occur between the time a signal enters and leaves the PT2399.  Here is my dilemma.  Do I assume Princeton's measurement accuracy is best at lower VCO frequencies (longer delay times), which is also the region of interest for tap-tempo stuff?  Or, do I use a simple average of all the data? 

Avg samples per unit delay (across all VCO frequencies) = 684350
Avg samples per unit delay (up to 9MHz) = 638190

[Andre]

I used 684000 as divider for the tempo indicator I did a couple of years ago and it seemed pretty accurate on different delay times.
Although I don't use the tempo indicator anymore I can connect it to my PT-80 and, in infinity mode, check how accurate it is over a longer period at
different delay times.