I've said it for a few years now and I still maintain that the only realistic way for RISC OS to survive is to move away from ARM based hardware to x86 hardware. JGZimmerle's suggestion is certainly one way of doing that. Using linux as a base certainly has its upsides but it probably has its downsides as well.
The future of RISC OS is always going to be expensive whether it's on new ARM based hardware or moving towards x86 hardware in one way or the other. Trouble is that the ARM based way seems to be far too much of a lottery - will enough people be willing to buy yet another "expensive and underpowered" ARM computer for the market to stop it's seemingly steady decline and start growing again?
I think it's too late already for ARM based hardware and x86 hardware is the only current way forward. It should be possible to do it in a similar fashion as Apple is doing - build a computer to certain specs and develop for that in the beginning (certain graphics card, sound card etc) but leave an opening for others to develop drivers for other hardware.
Backwards compatibility is something that people worry alot about, I think that's a mistake. Too much backwards compatibility has kept RISC OS underdeveloped for so long that it's killing itself. I still stumble upon software that is being held back just so that it can run on RISC OS 3.1, what on earth is the point with that? Of course some backwards compatibility is needed and always will but as has been proven time and time again does nothing but kill new developments. Why bother developing new software if the OS is being held back to support some software that hasn't been developed for a decade? Why should people switch if the new software can't offer any real feature improvements because the OS doesn't support them? Why should people even upgrade their OS when it makes no difference on available software?