I think I need a new router, but maybe I have another problem.

If I run a bandwidth test from a rb411 to a p4 1.6 running routerOS I get a full 100mb/sec Ethernet speed, but if I reverse it, I only get 38mb of speed and the CPU of the p4 1.6 is 100%. (Its test on a vlan with a switch in between).

那听上去对吗?

The traffic generation takes all the CPU cycles so it's hard to see if the router is burdened with packet generation or packet handling. You really need to setup hosts on both sides of each router and generate the traffic between those to get a real test.