I would not really say this is theoretical work - there is a lot of theoretical work on horn bends out there, but it seems to be mostly in the JASA which is harder to search or other even harder to get journals.

Anyway, looking at my FEA's again, I'd say that for a given frequency, with relatively small horn cross sections, large round bends work well. For larger horn cross sections, reflectors work better. If you're looking at the FEA's, the dimensions on the grid are inches, and as noted, the frequency is 5khz.

John