Poor timing and design decisions.
With Mediatek moving to designs with all big cores it's getting to the point where x86 isn't much of a problem any more for mobile either. As for Xeon Phi, CU/SM's in GPUs aren't small enough that using x86 for the scalar processing is a significant overhead ... it was how Intel decided to design the rest of the processor which made it impossible to compete with GPUs. Narrower SIMD, less SMT, strict coherence etc ... that killed the computational density, not x86.