ACM Queue has an interview with Steve Furber
about ARM and low-power computing - of course, he mentions the 6502 in the BBC micro, and what processor to move to next.
Rather than butchering a long quote, I'll paraphrase: they'd tried various 16-bit processors and found them wanting on two counts.
First, they had complex instruction sets, therefore poor interrupt response. The BBC Micro had no DMA, and needed fast interrupts for realtime I/O. They found the 16bit micros of the early 80's "had worse interrupt response time than the 6502"
Second, their models told them performance depended principally on usable memory bandwidth. Cache memories were "not commonplace, so the available memory bandwidth was determined by the performance of commodity DRAM." And yet, the processors they looked at couldn't saturate DRAM bandwidth. They felt that was "the wrong answer. The memory bandwidth was the primary resource, and it was the processor's job to make the maximum possible use of that. "
Then they "got wind of the RISC papers published by Berkeley and Stanford"