How much of an advantage are we talking here?Code density... It's still better even when you're strictly compiling ARM in thumb and don't have to deal with switching between thumb and full ARM instructions. The only comparable ISAs in terms of code density were M86k/Coldfire (which IMO is worlds better than x86 as an ISA goes) and personal favorites of mine, SuperH, and MCORE.
I doubt it can be a big difference, given that x86 will push and pop to stack relentlessly and needs to copy registers (vs RISC-typical 3-operand instructions, lack of registers, and having some idiotic stuff like forcing shift-amount in a specific register). I`d think that using Thumb or MIPS16 would come pretty close or even have higher density, but I lack any empirical evidence to back my guesses up . Do you have some data about it?