The question is: Where is the time going?

To get an answer, I tried turning off various parts of the file system object logic. Somewhat to my dismay, the extra bcopy within the file system application is quite costly. This bcopy reflects the absence of a scatter/gather mechanism in the underlying IPC system, and has pretty well convinced me to reconsider that architectural decision.

The ``write processing'' portion is really the cost of determining where the data is to be placed.

We'll get a fair bit back out of the basic IPC costs at these payload sizes, but not as much as I would like.

All told, we probably can back the factor of 2 needed with a bit of thought.