Well, it depends. If you need every person in the crowd to be individually controlled, then you should write a single surface system for the crowd of sprites. Handle frustrum culling yourself, or do the crowd as batches of people, with each batch a separate entity which Blitz will cull.
For a baseball videogame however (I saw your "Perspective" thread) you probably don't need that level of control since people in the crowd are sitting in rows. Furthermore, if the camera doesn't get too close to the people you don't need much resolution for each; even if the camera does get close occasionally, you could do a trick like cutting to a small section of high res sprites or even models. In that case I would do the crowd as a series of flats/sprites (make sure to set SpriteViewMode to 2,) with a texture of a whole row of people on each sprite. Thus you represent a huge crowd in the stands with a small number of sprite entities.
For a baseball videogame however (I saw your "Perspective" thread) you probably don't need that level of control since people in the crowd are sitting in rows. Furthermore, if the camera doesn't get too close to the people you don't need much resolution for each; even if the camera does get close occasionally, you could do a trick like cutting to a small section of high res sprites or even models. In that case I would do the crowd as a series of flats/sprites (make sure to set SpriteViewMode to 2,) with a texture of a whole row of people on each sprite. Thus you represent a huge crowd in the stands with a small number of sprite entities.