Some RISC processors are very fast (the PowerPC G5, UltraSPARC and probably the Dec Alpha (the first 64bit RISC processor) the latter sadly no longer with us), and some CISC processors translate their x86 code and have it run on a RISC processor core anyway. The deficiencies of CISC were to some extent addressed by the CISC processors being clocked considerably faster than their RISC compatriots. Pile on enough MHz and eventually you get some speed advantage.
That having been said it's dangerous to assume that 1MHz on one processor equates exactly (in performance terms) with 1MHz on a different processor architecture.
MS I don't believe are pushing processor speeds along, the hardware manufacturers would do that anyway. In the case of the ARM this was (once ARM was spun off by Acorn) retargetted at low power consumption uses, this meant using smaller memory caches, and not pushing the clock rates too high (and this hit performance).
That having been said the RISC PC managed impressive speed considering the very small cache size and lowish clock rates and that was as a result (in part) of the frugality of RISC OS and the inherent neatness of the ARM architecture.
It would be nice to see what the ARM would manage to do with a Pentium sized Cache and clock speed (I think when running RISC OS we'd all be very pleasantly surprised), especially considering that at the moment when clocked much more modestly the StrongARM and XScale can be as responsive at some activities as their PC compeditors.