Benchmarking is simple to say but much harder to do.
We all have to agree to a set of "common tasks" - impossible;
We will need to make sure that those compilers are not purposely optimized for those tasks - impossible once the set of tasks is determined and known;
We will need to code against / around potential optimization - very hard to do;
...
Having said that all, the limited amount of work I have done would suggest that speed-wise, GCC can give the commercial compilers a run for their money, if not beat them slightly.
Not a surprise to me: I use Keil / IAR not for their speed, or even small code size - even if Keil / IAR could beat GCC in those benchmarks. I use them for support (accountability?) and for certification (reliability / reputation?). I would venture that I am not unique in that regard.