The problem is, that even a 1GHz ARM offers only about half to a third the performance of a 1GHz Pentium or Athlon. If RISC OS could make use of multiple CPUs, the situation would be better with the ARM11, but it can't. Meanwhile, the Pentium has just achieved 4.25 GHz.
ARM's license fees are almost irrelevant, they only account for a very small fraction of the retail price of an ARM CPU. So it won't make much of a difference if ARM lower their their licence fees. The main problem is, that to get a good general purpose ARM-CPU for desktop use, we would have to manufacture it ourselves.
I was not proposing a simple Emulator on another complete desktop OS as the solution. What I am proposing is to use a Linux kernel plus a good JIT-emulator as a HAL, to extend RISC OS to enable direct use of the hardware's resources through the Linux-APIs and add an ELF-loader wich can execute application binaries wich have been compiled for the host processor (for example a !RunImageIA64, !RunImagex86, !RunImagex64 or !RunImagePPC). So applications wich offer such a native binary would get a speed increase of about 1200% over the Iyonix on current x86 CPUs.
A large chunk of the developing costs for new hardware goes into driver development. For the system I am proposing, only one driver would have to be develped for each type of hardware, wich interfaces direectly with the Linux API.
An example might be a SharedSound module wich uses the ALSA (Advanced linux Sound Architecture) API to drive just about all sound hardware available. Since ALSA is also multi-threaded, it can utilise multiple CPUs. Another example would be a graphics driver wich uses the OpenGL interface of the Linux drivers, so you could use any graphics card you like.
Because it would boot directly into the RISC OS GUI there would be no Linux-UI accessible to the user.
So to sum up, we could have a RISC OS computer now, wich is:
- 13 times as fast as the Iyonix for applications wich have been recompiled for it
- several thousand times as fast for floating point operations
- about StrongARM speed for legacy applications
- much faster than StrongARM speed for all legacy applications wich make good use of native versions of shared libraries and OS functions
- compatible with just about all legacy software, even ARM2 based systems could be emulated
- extremely fast for all I/O operations (HDDs, DVDs, sound, graphics, USB 1&2, FireWire, etc.)