As you noted there is no such thing and it's not easy to port an existing one over to ESXi, since it's Linux.
But even if there was, don't run benchmarks inside the limited ESXi base OS. You would probably get some very wonky numbers.
As a type-1 Hypervisor, the ESXi userspace shell based on a busybox environment is just a limited interface for management purposes separate from the actual vmkernel. It's limited in terms of resources and IO scheduling.
I know this might seem counter-intuitive at first, but benchmarking from a VM will actually allow you to make use of the full host's capabilities, since that IO really runs in vmkernel space from the host-perspective.
If you still really insist you want to test your hardware from a baremetal point of view, then boot the server into a live Linux of your choice and test there.