I'm wondering big the prisma overhead is here. I know that historically, prisma has been very slow - the sidecar process and the generic protocol and database agnosticism come with a steep cost, and the numbers shown in the benchmark both seem rather low to me. 4x the performance of "very slow" isn't super impressive.. but of course, I can't verify this right now since I don't have access to machine where I can duplicate this and run some raw postgres queries to compare.