We should write some benchmark that is checking the difference between building a block, importing a block and validating a block (calling validate_block as the relay chain) for a parachain.

The benchmark can be relative simple. We push as many transactions as possible to a block until all weight of the block is used. We may want to change the "benchmarking runtime" to allow 100% weight for normal transactions. For creating the transactions we can follow: https://github.com/paritytech/polkadot-stps/blob/main/docs/introduction.md
Maybe we want to split the benchmark into two, one compute heavy and one io intensive benchmark. For that we would need to select the best transactions that are either compute/io intensive.

There should only be a difference for the IO intensive benchmark as we keep the PoV in memory and read the data from there. The compute intensive benchmark should show almost no difference between building/importing/validating.