If its running to a breakpoint it should be really fast
I have tried it with single HLL step (one C line, not one asm op-code), just pressed F6 down firmly.
The st-link v2 has a 1800kHz clock (max) so there should not be a significant difference between Keil or OpenOCD + Eclipse. The difference might be that at each step an IDE and debugger has to interrogate target. The only thing that has to be made is to read PC, then move breakpoint to next position and then continue when doing such fast stepping. All the other stuff like reading IO registers or flash/sram is necessary only when an IDE needs to update ALL views (which is not needed when doing fast stepping).
So perhaps here lies the difference.
10 steps per second sounds really great, my setup does only 2 per second (which is still ok for me).