There are a lot of things I could say, in a followup to all the other messages but I'll just stick to some simple items.
SecondLife has grown a great deal since the simple little xml based system(that might have had some seed money from the Nez). For a long long time it's just been growing like crazy and when ever a problem occurs more often then not it's just patched.
1. It's going to be a slow and agonising business but the server code has got to be redone in something other then C++.
Misuse of C++ has been responsible for more f*cked up systems worldwide then anything else around. A proper programming team needs to be given access to the code and do something in another language. I appreciate what the "Firemen" have done in patching and other stuff but the server codes need a rewrite from the bottom up.
2. There are dependacies in the viewer code and server code that should be removed - most of the music and other media streams go thru the SL servers at some point in their journeys; that should be left to the users' systems once the server gives them the link to a stream.
3. Absolutely no new features added to the SecondLife system for at least 6 months to a year; mindlessly piling on new features again and again has damaged the system in many ways. The physics system is very impressive but it took a long time to get it working right with SL, same with mesh and other features. Linden Labs has a sick code base: that needs to be fixed first before throwing more things on to the system.
4. Work on Sansar should stop and the team working on that help with the main systems and make notes as they go. So they don't make all the same errors that original programmers of SL did - not planning for expansion, nesting many calls for resources so deep that DB machines often timeout on requests if the network is lagging even a little(i.e. Saturday nights). There are no doubt many other problems but if a good foundation is laid, the code should keep going for many more years.
Making SL and Sansar sisters instead of vaguly related systems would go a long ways to helping the stability of these virtual systems.