The current shared memory transport used by the new plug-in between the in-browser code and attached JVM instances is inefficient. Because we didn't have enough time to fully debug the "optimized" code paths, the amount of memory traffic and unnecessary signalling of the other side is very high. A transport using named pipes and minimal buffering should be implemented. This should both simplify the code and improve its performance. This may also immediately solve performance issues which appear to be present in the LiveConnect and printing code paths.