Performance!
Some of us that specialize in embedded work actually wondered just the opposite. For instance if you had two identically configured systems but one is MIPS32r2 based and one is an ARMv7 and both are clocked at say 500MHz, the ARM system nominally would consume ~15% less power than the MIPS system but it would also suffer from a ~25% performance hit when compared to the MIPS system.
Similarly I worked at a company that specialized in low power x86 SOC designs back in 1996 and I was responsible for doing an exhaustive comparison of the available low power architectures at the time. In my report to my supervisor I told him that Motorola's M-Core architecture would perform better than ARM on a power usage basis but not on a performance basis.
ARM had also implemented a programmer accessible shadow register set, special interrupt processing and the THUMB instruction set and the writing was on the wall.