Change the System
There comes a point where you can't optimize the system anymore. At that point you have to take more drastic measures.
The analogy a participant from Evolution Studios came up with at the XP 2005 workshop was: "When you optimize software, you profile to find the hotspots, you optimize the hotspots, you measure again... After a while you can't find any more hotspots worth optimizing. What do you do when you still need to go faster?" You refactor the software, use different datastructures, use different algorithms, maybe even re-examine the requirements.
What do you do when you get stuck optimizing a system. You try to change the system itself.
What you're up against is a more insidious, hard to detect bottleneck: the rules, procedures, processes of the organisation. Some of them are public, many are tacit knowledge, "the way we do things 'round here".
Changing systems is hard, slow and dangerous work. Donella Meadows has an excellent article on places to intervene in a system. One of the important points is to find the "leverage points", where a small intervention has large effects.