In general, I would propose using a half wave box (or sphere or cylinder). Lambda/4 is a little problematic in my view (too small). If possible, try (and error) with half wave box. Compare the results. If there is little difference in the S - Parameters (< about 5 %) it shouldn´t matter.
I had the same discussion with my professor, he also proposed half wave when structure is small and lambda/4 when structure is big. I somewhat share that view, but due to another aspect.
Seb