Add more variety in benchmarks

[bhouston]

Currently there is not enough real world examples in the benchmarks and it is distorting the optimizations I've done towards synthetic data workflows.