The output of show proc cpu shows 2 numbers, such as 20%/10%. The first (20%) is the total amount of CPU in use. The second (10%) is the amount used specifically by hardware interrupts. The second number plus all numbers listed below should equal the first number.

In your output, most of the CPU is being used by interrupts. The following link talks about troubleshooting high interrupt utilization. My first guess would be a problem with the cef configuration, so could you check the output of "show ip cef summary"?

Another likely candidate is if you have ACLs applied to the interfaces in question without enabling netflow. Depending on how they are configured, they could drastically impact performance. This is also covered in the troubleshooting page below.

http://www.cisco.com/en/US/products/hw/routers/ps359/products_tech_note09186a00801c2af0.shtml

What the other posters have referenced is more germane to software based routers. A L3 switch, like your 3750, should be doing traffic forwarding in hardware. The CPU should be used for only "control plane" stuff. However, there are situations where the 3750 will software forward (often from an overflowing TCAM - which sometimes indicates not using the optimal SDM template), device "attacks" (since you mention an Internet connection, this is more of a possibility), or some "bugs" in the IOS (your 12.2(25)SEE3 is very, very old, possibly the IOS that came with the device when new - if possible, try 12.2(55)SE12).