I have now had this conversation several times with people about the potential future that a mix of open source and open minded development may bring to the virtual world industry. Much of what is happening seems to be driven by some of the direction Second Life has taken or is taking, though not so much to follow it into corporate lockdown but to breakout and provide the flexibility and creativity that is needed for the next generation of virtual worlds.
In the early days of 2006 many of us said it would be great to be able to run a Second Life server, our own one under out control. That has taken a while to start to emerge, but it has emerged as an expensive product aimed at corporate IT departments. Luckily the opensource community had rallied and created the excellent Opensim. This ticks all the boxes of being able to be run locally, be run in the cloud, be provided as a service. So we have an extensible virtual world server ready to be built upon.
The other component missing was a more controllable and rich interface. Yes there is the Second Life Snowglobe open source client but the need to certify and lockdown variants to align with the product needs for Second Life means that lots of the flexibility is lost. Likewise the initial open source Linden based client was under a GPL licence which caused all sorts of development to not happen at the time it really could have done with it.
This is where Unity3d steps in I believe. It was Rob Smart who first started to show me this way back. Unity3d is a great front end, very flexible in how you build games and content for it.
This was a movie form back in September 2008, using a message from Second Life to several unity clients to create a cube. This is loose integration, telling one place something has happened and letting the other place get on with it.

Unity3d has a plugin architecture too. It runs in a browser or deploys to application platforms like mac and windows. The visuals can be made very good very quickly too. Unity3d needs a server of some sort to operate as a multiuser platform (though it does do some peer 2 peer) hence applications like Smartfox are ideal for producing Unity multiplayer and MMO style games.
However Opensim has all the other layers of things needed to maintain a virtual world. It has assets databases, chat, positional awareness, server side scripting (as does Second Life that is was originally based on of course).
So we have an extensible and easy to get hold of Unity3D client engine, and extensible and easy to get hold of Server/Persistence VW engine in Opensim. There may well be challenges in making the two understand one another but with the flexibility both sides of the equation that makes them very solvable. This is a high level view, Rob has some more detail here on the challenges. Add in some interoperation definition with Vastpark to help bind the two and make some mappings.
Throw into the mix an open minded avatar wizard such as Evolver. There we can build avatars that we know definitely can be dropped into Unity3d.
So…..I create an Evolver avatar, dropped as a resource bundle into a web deployed Unity3d client that tells the opensim server where I am in the coordinate system, and which bundle I am using. Other people with a Unity3d client see the rich detailed avatar and the shiny Unity3d environment. However we do not have to stick to that one client. Other people using a Second Life style client see the Second Life style rendering of the world?
This is already happening in some respects, the Iphone application Touch Life lets you logon to the public Second Life. In a sort of bugblatter beast of traal moment everyone can see you, but you cant see them. You navigate your avatar around the map, have full chat, inventory and economy access, but a very different view of the world to everyone else. (Of course Unity3d runs well on an Iphone too, so imagine that as an extension to Touch Life?)
Once there is an acceptance that there can be more that one view of the data, one where people without the full equipment can still see what is going on and participate things get a lot easier to consider. Whilst a gaming assumption tends to be we all need the same view at the same speed in order to be able to have balanced gameplay (lag gets you killed) in collaborative spaces, education, meetings and art galleries this is less of an issue.
As the parts of the jigsaw come together over the next year the ability to have the same experience will re-emerge.