I have created a simpler version of the test. (it is
easier for me to follow, anyway :) Attached are
my .orc and .sco. (can I use a .csd with the
vanilla version of Csound? Sorry I have not kept
up with the Csound developments!).

I agree with Gabriel - Gen08 is doing a better job - 
it seems to be, anyway, just looking at the output.

There seems to be two problems in fact, but they may
be related. Not only does the shape of the oscil3
output look worse, but oscil3 is not reading 
back at exactly the right sample rate. The way
I have written this test, oscil3 should produce
alternating interpolated and non-interpolated 
values, and the output should repeat for every
cycle of the sine wave. This is not happening - 
the output samples very gradually get out of sync
with the table. This can be readily observed by
loading the output into a soundfile editor, and
noticing how the overall amplitude of the signal
appears to drop slowly over time.

Gen08/oscil is not exhibiting this - it stays
in sync. 

Just as an aside, Gen08 does of course generate
"bad" values at the end of the ftable, because
it doesn't interpolate using wraparound. So
if the output is listened to, it sounds dreadful,
but this is not indicative of it's interpolation
quality. To get around this problem (to a small
extent) I gave it a few cycles with which to work
with, which produces a cleaner, and more representative,
output.

Greg.