I'd personally slow down and pitch correct the PAL audio out of the two, since the FILM length is what it was originally meant to be watched at and the PAL version is only faster due to standard constrains. One or the other would have to sound somewhat worse even with pitch correction, I'd rather it be not the original personally, but in the end I guess it's up to you and there isn't a "better" choice, so to speak. This should cover both the first and second question of your latter post.
As for the other things:
- Depending on what you do, the results differ. No way is perfect but here's the ways one can employ to get from FILM to PAL:
- Speedup directly from 23.976 to 25. Here the frames will be intact, but the alteration of the playback speed could be noticeable and awful depending on the content and the audio. For instance, based on how the background music and sound effects are treated or if there are already mildly fast scenes, things will get bad in motion. Audio side, because one can recognize the speedup (especially true for older works back when they didn't even bother with correcting the pitch), video side, because sometimes there's a person moving at an unnatural pace for the movement they're doing and so on.
- Speedup from 23.976 to 24.00 and telecine. This could be hard or soft telecine. The speed difference is slight enough that it will hardly be noticeable at all, especially with proper pitch correction as far as the audio goes. The downside is that of course motion will slightly stutter. However, if it's a soft telecine and the device doing the playback is able to ignore the pulldown flag, it'll just get played back at 24.00 fps, so it's going to be smooth too. Soft euro pulldown is quite possibly the least bad solution available.
- Speedup from 23.976 to 24.00 and then adding a dup frame. Compared to the above, the motion stutter will be more noticeable. OTOH it will be progressive, so there are reasons one might want this.
- Blending everything to 25 fps. The most common choice nowadays. There won't be issues with the audio, which is a plus. However, video quality will suffer. This could happen either progressive or interlaced. In some rare occurances, with interlaced blending, you might be able to restore a clean stream if there are clean fields still in there, with progressive blend you just can't.
- The answer to this is included in the former question. Or, well, I assume that by NTSC you actually meant FILM. If it's actual NTSC to PAL, then the easiest solution is probably to just drop a field in every group of 6. Could have a slight motion stutter but at the end of the day with that amount of fields it could possibly go unnoticed.
- As far as avisynth goes, you can do a hard telecine by combining separatefields(), selectevery(), and weave() (some simple logic will help you get the parameters you need for selectevery). You'll also need assumefps obviously (to bring 23.976 to 24.00 before doing the telecine). For soft telecine, you just set the euro pulldown flag when doing the encode after just the assumefps.