Testing this locally resulted in improved response times from ~50ms to ~20ms when creating a project. It looks like even though the code running these sql queries uses different go routines, they affect each other (caused by IO or context switching?)