It is worth pointing out that both of these programs are written in BASIC (runs from CPU cache) and RISCOSmark is designed very much to fit in the CPU cache. As such, you have to be careful when using these benchmarks on different CPU architectures. Translation - programs like this aren't overly representative of real applications, so handle with care.
RISCOSmark scores will vary with screen mode (it runs in desktop screen resolution/colour depth) so one has to be careful comparing (some machines will run faster in higher colour depths, for example - I wonder if the A9 is optimised for 32k or 16m colours, as 800x600x256 isn't exactly "real world"). When I tried them here, I rather liked Firebench although that may be because my RISCube did 1046 fps! On the other hand, the CPU rating in RISCOSmark was chronically low - 130%. Neither really represents the "real use" speed of the machine although it is closer to Firebench than RISCOSmark.
Obviously it is pleasing to see the A9 scoring well, and no doubt they will improve as more acceleration is implemented. Even so, they're already impressive!