These objects were primarily designed for generating accurate sub-
sample values for reading from large buffers, and filters where 32bit
precision weren't adequate for filter coefficients. If your FM
algorithm contains feedback, you might find that these objects
provide results with less accumulation of quantization error, though
whether or not that is even a problem or sounds "better" is obviously
dependent on what you're doing, and what your ear prefers. I would
recommend using the standard objects first, and then perhaps
experimenting with the hr objects to see if you can notice a
difference. For FM, I'd imagine you won't notice a large difference,
however I'd be interested to hear your findings.
I think the sample rate would make a big difference for
high frequency FM. The higher the sample rate, the fewer aliased frequencies appearing in the audible range. I can't comment on
quantization error or how much difference it would make.
If I were you, I'd forget FM and just go with Phase Modulation instead. Once you start builing complicated FM stuff (multiple carriers/modulators interacting with eachother), it starts to sound like crap (imho), especially with feedback. I can't give a DSP math explanation as to why. Russel Pinkston wrote in the Csound book that complicated FM can introduce DC offset (I can vouch for that) and carrier drift.
Here's an example of 1 carrier, 1 mod PM with feedback:
i had a mess around with replacing all the FM with PM
in my synth patch, and it sound sharper, but not
necessarily better. for instance with FM you get a
pitch wobble on attacks as the oscillators drift from
the fundamental, but that actually sounds sort of
realistic. the PM sounds alot more synthetic to my
ears, which i guess could be good sometimes.
i'd be interested if any one else has experimented
with PM versus FM synthesis