I took another look at the group delay, it's worse than I had remembered.
I updated the diagrams above. About 250nsec for luma and 300nsec. However, the spectrum that would now be chroma only gets some group delay predistortion and thus it still looks okay on a TV set.
My other filter's frequency and group delay response:
And the circuit:
It's a tradeoff between frequency and group delay.
You'll have to decide which looks better.