A few basic questions about your test methodology...
- What platform are you testing?
- What test tool are you using and how many simultaneous connections is it opening?
- What does your virtual server configuration look like?
- What is the CPU usage on the LTM during the test?
- Is anything logged to /var/log/ltm during the test?
3Mbps is slower than any performance result I have ever seen other than during horribly broken tests. The only time I could see this being an accurate result might be when the test tool is opening only a single connection during the test, and you are using a virtual server profile meant to optimize features over performance. (Though even then the key part would be 'single connection' rather than the profile used on the virtual server.)
--jesse