Short answer? Consistency, repeatability and price.
By the nature of it's composition concrete contains much larger, less stringently filtered aggregate components. Therefore despite having relatively uniform surface characteristics in any given sample sized to reflect it's usage(for example, 1 meter x 1 meter to simulate it's usage as a sidewalk material), in samples used for this particular application one would see a drastic variance in surface characteristics. So whereas you may see in a panel of ten 1m x 1m samples that on average 1/8 of it's surface is composed of exposed pebble and 7/8 is composed of sand and cement compounds in a panel of ten 5cm x 5cm samples you may encounter samples whose surface is composed almost entirely of exposed pebble. If one of these samples were used in testing the results would vary wildly when compared to those whose surface structure was almost entirely made up of sand and cement compounds.
Then why not exclude pebbles altogether and make your samples exclusively from Portland cement and sand? The exclusion of the pebbles changes the cohesive properties of the concrete structure so your surface, even using the same exact sample repeatedly, would change with each testing as the surface wears away from repeated strikes whereas sandpaper glued to steel offers the same relative hardness and roughness in every test. I do wish they had used a courser grit, such as 36, to more accurately simulate concrete's rougher surface though