Hashing often has much better constant factors than binary search because it isn't super mean to branch predictors
10, in particular, just feels odd. That many items almost surely fits in a single cache line.
Granted, if you are storing compound data such that each item is a reference elsewhere anyway, I can see this. Speed is likely dominated by the comparison procedures at those sizes.
That is, an unrolled loop on bytes can be N conditional increments on an index, with a check at the end to see if still in the bounds. Assuming N is small, will be hard to compete with that, honestly. The hash would be spreading the data wider than the linear search would be. Though, I agree I'd expect it to still be on a cache line for bytes.