> It is moderately trivial to modify gcc to avoid the bug, however there
> is a cost and this potentially means maintaining two versions of every
> binary, which we just don't have time to do.

Then don't use separate versions. 

As I understand it, all gcc has to do, is to put an extra NOP
before the LDM or LDR instruction if the instruction-address is
on a page boundary. This should make the code runnable on all
ARM processors. 

Considering the size of pages, and the frequency
of LDM/LDR instructions, I get that the programs will get bigger
and slower on the order of  1/10 000. 

I think all can accept such a small penalty.
It is definitely easier to accept that, than to accept that
most RiscBSD users has to change processors.

Kim0