I always assumed that compilers for CISC were easier to write and have better performance because the instruction set was much more high level than RISC. For example multiplication is a single instruction in x86 while the RISC approach takes four.
Nope, it's easy for a compiler to spit out four instructions instead of one. It's much, much harder to deal with the typically non-orthogonal register usage of CISC designs, not to mention that optimising around big complex instructions is actually much harder to do.
x86 won out despite everything, including Intel's various attempts to kill it, because there is an enormous amount of x86 software out there that people need to run.