for a given overlapping photos footprint you could choose a range of lenses. As long as focal lenght increases, the distance to object must be longer to have same coverage. As long as you get farther, the two images will become more similar one to the other.
Notice that this might carry advantage in terms of correlation: the more similar appear to be the feature points, much more successful matches will be found for that pair... and that is great BUT at the same time the projective corresponding rays will intersect at smaller angles yielding to poor accuracy in 3D solution specially along the z axis (from object to camera). And that is BAD
On the contrary, as long as lens get wider, ray intersections become closer to right angles. That in terms of accuracy is great BUT images become more different one to the other, this means that fewer feature point matches will be found because of their poor correlation... and that is BAD again.
The sweet point is somewhere in between wides and teles and this might be the reason to recommend normal primes.
In fact there is no exact rule but in general I'd say for aerial works you can go directly to wide angles in most cases. If there is not too much vegetation, and the terrain altitude range is not extremely long do not hesitate to use 35 or even 24mm equivalent lens.
For object scans (convergent pictures), I would recommend longer focal lenghts because the quality of the intersections can be ensured by adequate camera attitudes.
GEOBIT (Agisoft.es)