A clever trick would be to cache all options in the 8-second voting window and then blast them out to the clients. By the time the vote concludes, the clients would already have the image of the next step, and the server would have had 8 seconds to send that to 400+ people. Plus, only a few maps API calls.

Though I'm not sure the StreetView API allows caching. Having worked with maps APIs in the past, I know they often don't allow it.