If your using several objects, then you'll hit a brick wall of performance issues. I think though, that your half-million polygon count isn't the issue - as Bendy says, the number of draw calls would cause a problem.
I would suggest making a crowd of spectators - like say 10 x 10 spectators, and randomly welded to each other. So you might have 5 limbs on your object, but that's the spectators glued together, so you offset limb 1, and it'll offset more than one spectator - if you offset on Y with an absolute sin value, the spectators start jumping up and down, but in their limb groups. That's assuming you would want them to be able to jump.
If you can, you should reduce the polygon count the further back you go - even inside that group object, maybe front row, each spectator gets 400 polygons, second row 300, third row 200, fourth row might just be billboard, single polygon plains as Brendy suggests. But this way, you would be able to save a lot of performance, greatly reduce the number of objects, allow for wasted polygons in crowds further away. You probably don't even need to make the whole crowd - just fade them out to darkness, but have random camera flashes from all over the arena - I think you'd end up with a good crowd effect by doing that. Even in the most popular, high budget sports games, crowds are often pretty terrible - often to the point of incompetence I'd say

.

I got a fever, and the only prescription, is more memes.
