I am working on a Particle System, since current graphics cards prefer to have large batches of data rather than many small batches I want to reduce # of batches. There are two ideas that I am toying around with: shader instancing (my hardware is not capable of preforming hardware instancing) or using a dynamic vertexbuffer. Shader Instancing has N (N=255) constants and would require no locking of the buffer, but would only allow N particles per batch. Using a dynamic vertexbuffer would mean it would have to get locked but would allow for fewer batches. It would then be a matter of locking versus batches... I do not know which one to method to use, if there is a better/different way I am all ears.

I think you should try both ways and time them. The results probably depend on the number of particles. I would be interested in knowing what you get.