Surely not an expert, so only @koconnor can give a definitive answer but I would assume that “real” printing is quite different to synthetic benchmarks.
If I’m understanding this correctly, the benchmarking will just toggle n number of pins as fast as possible without loosing the required timing precision. Surely sufficient to compare the performance between different MCUs.
In real printing you will have a lot of additional tasks, like streaming / processing the step queue, measuring / controlling temperatures, exchanging relevant information with the host etc.
Don’t quote me on this, just speculating.