It's been a while since I was in a position where I regularly watched servers boot, but 2015ish I was doing a lot of 'boots on the ground' sysadmin and virtualization work. Servers do a lot of additional testing during boot-up, temperature sensors, RAID cache batteries, memory and RAID arrays are all checked. Some of those checks can be disabled, but you don't really reboot production servers regularly so you typically wouldn't want to. The extra 3 minutes of boot time is much easier to deal with than a bad host coming online.

On top of all that it's pretty standard (at least in the VMware world) to store the OS on an SD card. So the OS has to be read into memory and ESX is kinda slow to boot even if installed to a disk.

Yes. It's just not a priority for servers.

The reason it takes this long is that it does a bunch of self tests, and then it has to load all the ROMs for the components (NICs, HBAs, etc.) which often trigger messages like "X Loaded, press CTRL+L to configure" which stay on screen for 5-10 seconds each.

Yes and no. There is a longer boot time because the POST process is a bit more intense. ECC RAM is a big part of that. Chances are this particular caes is likely becuase the boot order needs to be reconfigured and it is hunting for PXE/network stuff before hitting a timeout?

If you yield control to some kind of Broadcom controller it'll do all kinds of shit before giving up and handing you back to boot a disk.

I rebooted a series of Oracle SPARC machines a couple of weeks ago that took about 20 minutes to boot/self-test.

Servers are built to be reliable. It is better for them to boot slowly and correctly than quickly with silent memory failures.

Unfortunately it is quite common that servers are slower to boot than desktops if it has a lot of CPUs, or a lot of memory, or just a very slow BIOS impl that spends a lot of time probing and initializing everything. Servers with hundreds of cores are a lot slower to boot than my laptop too.