I don't know if the filter will produce better results than what I've done manually, but it sounds like better minds than mine have been at work on the issue.
Edit:
This is restore24's method... does this sound wrong to anyone else?
1. Bob the video to 50fps
2. Detect fields that are blends
3. Replace those blended fields with unblended neighbors
4. Decimate the cleaned stream to 24fps
In my mind that means he separates the fields... I was doing that.
Detects the blended fields... seems to be the odd field only. Was doing that too.
Replace with unblended neighbors... why exactly?
Decimate... This only makes sense if there wasn't a speed up... which I can't confirm or deny but it sounds like a percentage of clean unique frames might get discarded.
Can someone give me the final runtime of the BBC version (I only have the first VOB)? That should indicate if there was speed up or not.