If (some of) the bots don't know they're not human, how do the humans know?
I'm going to get to the government in just a moment, but this is important. I suppose the software could let the actual humans (and the software would need to know which are humans and which are bots) could let the humans see the bot avatars with a big honking sign that says "I'm a bot!" while the bots, looking in a VR mirror, don't see the sign — but your stronger AI bots would start figuring something out even if you did that.
My point is, in the long run, there's likely no way other than through serious interaction for bots and humans to figure out who's who — and you had better hope they never do, because the one and only way to enact government will lead to a bot-vs-human war that would make WWII look like a couple of school boys having a pissing contest.
Governments operate on the threat of violence
There's no such thing as a nice government. There can be effective governments, possibly even efficient governments, but there's no such thing as a nice government — because the only option a government has to maintain order is to take something away from the governed and you can oversimplify the issue by reducing tht "something" to two things: resources and liberty.
Seen through this simple lens, the only difference between a government and a street gang is this: one is considered desirable and the other isn't.
What sort of government would be up to
you the players
Any form of government that has been tried on Earth could be enacted in your VR environment. In fact, I suspect if you visit places like Second Life you'll discover a number of governments operating in different locations. Where there is an economy, there must be a government. However, in the case of something like Second Life, the government is inherently overseen by the Terms & Agreements of Second Life.
And here's where you get to make a decision. It's your software. Unless you open source the software, the software owner has complete control over the environment and can impose restrictions. In such a case, without serious hacking (see narrative necessity), you know exactly which bots belong to whom and it's trivial to impose a one-vote-per-user limitation.
The problem is if you open source the software or otherwise free the system so that only an in-VR context can be known. Why you'd do that is beyond me, software (like any complex system) and the hardware that supports it must be maintained. It's very much impossible to "set it free" and make the system wholly non-interactive with the "real" world.
Which brings us back to the threat of violence.
What can you do to a player/user who refuses to comply with the rules? You can take his/her resources or liberty. Everything from "you can't visit this location for an hour" to "we've permanently deleted your account and the $8,524.23 worth of enhancements you paid for are permanently lost. Obey the rules next time."
Before you develop your in-VR government, you need to figure out what the relationship is between the real world and your VR's users. That relationship is the "actual government" because it imposes the harshest restrictions.
After that you can develop your in-VR government, or let your characters develop it, usually in support of whatever economic system has been imposed by the VR's design (remember, you have inside-VR money and outside-VR money, they will touch, and that means your in-VR government has something to do with not only the T&A of your system, but the very real governments outside the VR, too.)
And if you're lucky, your smarter bots won't figure out that they have power to influence the "people" around them, imposing the same mechanics of violence that you permit for the operation of the government to get their way.
Vive la Révolution!