He appears to be referring to the writing of benchmarks. Specifically for those in which functional programming excels over imperative (for example undo support), it becomes a significant undertaking to write the C version so it just doesn't get done. This means that there is a selection bias towards benchmarks that can be easily written in C. These simpler benchmarks tend to lack any need for the higher-level constructs that functional programming provides, which FP advocates would argue is exactly where FP gives the greatest gain.

edit: slightly less snark.