I have created a python-based benchmarking suite to compare the GRAS and GR scheduler, as well as implementations of blocks in both GREX and stock GNU Radio.
The performance is measured in BogoMips - that is bogus million items per second. The execution of a finite flowgraph is timed -- The throughput, items per time take is used as a performance metric. The actual numbers are not truly meaningful (hence the bogus part). However, the relative comparison between implementations is valuable.
Many of the benchmarks attempt to compare schedulers side-by-side, as well as benchmarking the overhead of the same block, but rewritten to take advantage of GRAS's features. For the comparison benchmarks, both GRAS and both GR default to 32kiB buffer sizes for "computational chunks".
The query client showing side-by-side overhead comparison using a pie chart: