[galaxy-dev] Single job latency

This Galaxy speed and production question reminded me that I have been meaning to raise the following question on the list:

For a single job (workflow item), It seems like there is a 20 second lag before the job is launched. It gives the impression that Galaxy is slow and unresponsive (in the sports car sense). I imagine that it is a combination of python's poor thread architecture and web polling latency. Has anyone looked into the latency of running a process?

Ted