I had a discussion with a few people about this at the MUM, and basically what people told me was that the bandwidth-tester is working great as a receiver of the test, but the actual package generation is too cpu intensive. I have a few suggestions on what you could do:
1 (and most prefered): Release a bandwidth-test client that is either cross platform or has a linux and windows version. That way I could load it on my different management servers, and test against the CPE's. Despite the low cost of Mikrotik routers, it's simply not practical for me to set up a dedicated RB for each VPN customer (my customers are Layer3 VPNs, and are protected from each other).
2: Change the testing algorithm on the clients.
At the moment, the actual bandwidth test does a lot of smart tests with different packet sizes and all that. If you made it possible to do very specific test, for instance by selecting what packet size to run (together with MTU, buffer sizes etc.), you could basically just cat all ones (or fill a repeating string, like ping does) into the payload of the packets and thus save a lot of CPU.
I know there is a bandwidth test program for windows, but I simply refuse to install wine on my servers, and run a performance test on an emulated/reverse engineered OS stack...
I hope you will consider my suggestions, and I regret I didn't catch one of you MikroTik guys at the MUM and discussed it with you there, when I had the chance... I blame the beers
The road to hell is paved with good intentions.