Since about 1992, ARM has always been targeted at the low-power, embedded market. The instruction set for many years was kept intentionally simple to keep the transistor count down to save power. The trend in desktop processors seems to be the opposite (more features in the instruction set). Also, it can't be stressed enough that people use whatever instruction set gives them the most access to software, and x86 has been by far the most dominant ISA in that regard. PowerPC, SPARC, and Alpha are probably the next 3 as far as software availability is concerned. In the embedded market, supporting legacy code hasn't traditionally been as important because there simply isn't that much code to begin with. And since ARM
is the traditional instruction set for many embedded applications, it has an advantage there. It hasn't been until very recently that people have seriously considered running x86 code in embedded systems (e.g., Transmeta).
As for Alpha, I just read recently that Intel acquired the rights to the Alpha from Compaq. I don't really know much about the Alpha's instruction set architecture, other than it is a RISC architecture similar to MIPS. To be completely honest, Alphas had very good floating point performance but were NOT Mhz to Mhz faster than competing processors. The Alphas managed to get higher clock speeds (they beat everyone to 500 Mhz back in the day) by creating huge multi-stage pipelines, similar to what Intel is doing right now with the Pentium 4. They WERE "faster" in SPEC benchmarks than their competitors but not by a margin suggested by their clock speed advantage.
Note: the above statement regarding the relative performance of the Alpha is MY opinion, and is no doubt controversial. So please don't state it as fact when arguing with others about the relative benefits of different instruction set architectures unless you like living on the edge
