As others have said, this is a huge area for discussion.

There is no magic bullet that you can apply and suddenly get improved performance with reasonably little effort involved; your example of updating to a newer SDK will be much more likely to bring you bugfixes instead. Upgrading to a newer D3D version is going to involve a huge porting effort, and if you do a naive port you're going to end up putting in a LOT of work only to end up slower.

You're going to need to get your hands dirty for this.

What you have to do is get a known good profiler, profile your application, determine it's bottlenecks, and target optimizations at those areas. That on it's own is a big enough topic, but you can get some rough ballpark guesses if you've set things up to be reasonably modular; e.g. by commenting out your particle drawing calls and comparing times with and without particles you can get a feel for how much time you're spending in your particle system (this is just an example; things may be different in your own code). Do this for other subsystems of your renderer and you begin to get an idea of what the overall picture looks like. PIX is also a good tool for helping you know where time may be being wasted in a frame.

I'd also recommend that you read the MSDN article "Accurately Profiling Direct3D API Calls".

when you've done that, maybe come back with some more specific questions relating to individually identified bottlenecks and you'll be able to get better advice on how to tackle them.