However tempting, I would recommend not using ever more than 1-executor agents.
Especially for heavy tasks related to development.
In $previousJob we've burned hours of engineering time to understand what was causing random failures when the wrong jobs would run together at the wrong time.
TL;DR it's mostly a matter of machine load, and it's exponentially easier to have more agents, isolated with a resource limitation layer like docker if you run them on the same host.
And if you're using the cloud, just set up two smaller VMs or so instead of a single big one.
Cheers