It also serves to show how well other implementations perform: CLISPc, which is based on bytecodes, is astonishingly fast, compared to other bytecodes implementations (ECLi, CMUCLi) and to native code implementations (ECLc, CMUCLc).
All benchmarks are executed one after another. The benchmark currently shows just mean execution times in seconds, using English decimal point (0.03 = 3/100 seconds). This means that performance of garbage collectors has a deep impact in these numbers.
It would be nice to collect some other information, such as memory use and garbage collector timings, but that is beyond my reach for now. Help is always welcome.
Finally, I have decided to distribute Gabriel's benchmarks together with ECL. The reason is that this code seems to belong to public domain and I had to perform several changes in order to let it run with different implementations. To recreate the benchmark just get in build/gabriel/ and type make.