Thanks for your reply. That's not really what I'm looking for though. I'd like to leverage the inbuilt alignment properties of the LabelClass, because:
1. Spatially positioned text labels don't work at multiple scales (especially a problem when printing), and
2. Our interface is crowded enough as it is. The audience for this tool are not GIS professionals, so I need a generic solution which requires little to no thought from the user.
They need to be able to check a box which says "Label this", and the system needs to make a good, educated guess as to how to label the graphic. We do that with Dynamic Layers on server-side & it works remarkably well. Unfortunately, users cannot label their drawn graphics without resorting to a new graphic of type TextSymbol, and spatial positioning.
We do use local storage for a lot of this stuff actually, but my performance concerns are based around adding more and more layers into memory. We have some Citrix clients, and this is a real problem for them.