I have Linux VMs which boots in ~2s (to ssh login; ~700ms to userspace). I use systemd-boot because it has nicer integration with my distro of choice than bare EFI stub, and it adds very minimal overhead in boot time.
Use linux-image-cloud; use a modern hypervisor with no legacy devices (only virtio etc); use tiny-initramfs; mask systemd-networkd-wait-online.service; add cryptomgr.notests noreplace-smp to kernel command line.
I think that's all, and use a moderately fast CPU (i7-8700K which is already 5 years old).
When you say modern hypervisor, do you mean, not KVM? How do you get around the UEFI/BIOS post time which is usually a few seconds by itself?
I'm talking about the time taken by OMVF/SeaBIOS or whatever equivalent in the VM; which is why I asked about the type of VM/Hypervisor because that might not apply.
I noted recently that it also sees systemd-boot installations and will boot from those, if desired (in my case it was a fresh installation of Pop OS, which uses systemd-boot by default).
For me, it's unambiguously better (meaning easier, with no extra periodic steps required to maintain my ability to boot whatever bootable thing I want), but I am curious about what scenarios would make one want to use systemd-boot (or grub) instead.