JGZ: oh dear. No, you didn't check at all (this is a common theme with you). You didn't offer any evidence or justification for your claims, and you don't seem to have run any benchmarks.
All you said was that it was "significantly faster", which clearly isn't true.
Futhermore, had you researched this a little, you would know that in fact, some of improvements for StrongARM cause XScale to slow down, and vice versa. GCC's default output is targetted to ARM6 level, which means code runs sensibly on all RiscPCs and later machines.
I'm not even sure how much difference targetting to these other processors will make on RISC OS - I do have some figures in mind from other OSes, but again, it's not something I will suggest without actually checking.