The performance is really good but the main point with the algorithm is the simplicity and versatility. It work also for world/view space aabb and child nodes. It's also does not depend your camera code. Also its more accurate than AABB methods. For static objects you can calculate world space AABB and save matrix matrix mul.
For special purpose case like voxel engine you can even optimize it even more. You can use matrix that include voxel size and center offset from 3d index so you precalculate extent part and result is frustum vs aabb test that cost same amount than normal frustum vs point.