Pete,
The reason I used multiple cycles in Gen08 is that I wanted
to compare apples to apples. Gen08 does not, to the best of 
my knowledge, interpolate with wraparound - it treats the
ftable as a linear list, instead of a cyclic list. Naturally,
we want to test the  interpolation for continuous tones, 
not a single cycle. Oscil3 interpolates the table with wraparound.
So, I used multiple cycles, and when I did the spectral analysis, 
I used the "stable" part of the waveform, without the yucky things that
happen at the end 
points. I believe that my test makes both oscil3 and Gen08
do exactly the same amount of interpolation - as far as
practical, anyway.

But yes, I did make a mistake and leave off the end point.
That wouldn't change the results I don't think, because I didn't
use that part of the waveform anyway. If anything, it would
only make Gen08 even better than it already is.

Seeing that Gen08 is performing better than oscil3, I agree
with Gabriel that oscil3 (et al) should be examined closely,
to determine why this difference is occurring.