Dead A3000D - not even the LEDs are turning on!

[JoeMuc2006]

Checked _KBCLK (green) and _KBDAT (yellow) with the scope, and compared to an A2000 it looks like _KBCLK is being pulled down by something in the faulty A3000. The keyboard event is in both cases "Q down" so keycode $10.
Sorry, the timebase is not the same in both screenshots but what is clearly visible is that both signals start high and return to high in the A2000. The A3000's _KBDAT is similar but the same does not apply to _KBCLK which idles at around 1V instead of 5V even outside of keyboard events (probably that's too little to count as high). It reaches the low state properly though.
The connection between the DIN terminal's _KBCLK and FAT GARY pin 44 is fine, measures less than 1 Ohm.
I measured the resistance to ground from _KBCLK without power on, and found that it's ~1.5kOhms when FAT GARY is home, and open line when it was pulled. Curious. Might be another thing pointing towards GARY being the culprit? However, 1.5kOhms is a little too big to pull the signal down that much, right?

[Castellen]

There's definitely something wrong with _KBCLK in your second image, as the line should of course be around 5V while inactive.  That likely explains why the system isn't coming out of reset state.

Have a look at the schematic: http://amiga.serveftp.net/Schematics/A3000_schematics/A3000_R9_schematic.pdf
There's not much connecting to _KBCLK, so you can remove the socketed ICs that connect to this - U110 Gary and U350 CIA, then see if there's any change to the _KBRST condition.

_KBCLK has a 3.3k pullup to 5V, meaning that it should be at 5V with U110 and U350 removed, and no keyboard connected.  If it's not at 5V, you may have to inspect where the track runs on the PCB to look for some kind of physical damage.  Maybe try removing C421, which is just performing EMC decoupling - the circuit will work normally without C421 fitted.  Don't rule out some kind of contamination in the 5-pin DIN keyboard socket either.

[JoeMuc2006]

All right, one step further.
The ODD CIA (U350) is at fault as it seems. Removing the decoupling capacitor for KBCLK didn't do anything but after I swapped U350 and U300, the reset logic works as it should. So if the ODD CIA is actually broken, it might now affect the parallel port but as that port isn't going to be used anyway, I think we can live with that. The keyboard was behaving as it should at first but now it's only briefly blinking caps lock about every two seconds, and not reacting to anything. I'll have to check the signal levels again.
The next issue now is that the board won't boot when a SCSI disk is connected to the internal SCSI bus and no FAST RAM is present (but 2MB CHIP). It's giving me a guru-style red error display with code 87010001 which points towards dos.library not having sufficient memory. I don't remember this for sure but it might be totally normal that using SCSI devices requires at least 1MB of FAST RAM installed. It looks a lot better after I moved the optional second MB of CHIP RAM (DRAM DIP chips) to the appropriate FAST RAM sockets and configured FAST to be 256Kx4. There's no more error now and the HDD can be seen in the boot menu. (not bootable though because it contains only data)
Overall, it doesn't look too bad now. Thank you @Castellen for all your advice, it was really helpful!
I will give another update shortly once anything new has come up. We are planning to build some ZIP2SIMM adapters to use the more convenient SIMM modules. PCB order was placed and module slots are at least available. These adapters have become really rare lately. I'll give a shout if we have any excess units to sell eventually.

[Castellen]

The serial port in CIA U300 isn't used, so if that's the only defect that the 8520 has, you should be OK with just exchanging the position of the two parts.  Keir's Amiga Test Kit provides some easy CIA diagnostic tools to quickly see what's functional or not.  Write the provided bootable disk image to a floppy disk, then you'll just need to make yourself a couple of simple DB25 loop-back connectors for I/O testing, the details of which are included in the CIA diagnostic menu:
https://github.com/keirf/amiga-stuff/releases

I don't recall any specific issues where the A3000 SCSI interface needs to have fast memory to operate.  It should still operate with only 1 or 2MB of chip memory, though booting will be slower.

Check that 5V termination power is present on the SCSI interface (pin 25 of the external DB25 connector) as some devices need it for bus termination to operate correctly.
 People often like to plug Centronics printers into the external SCSI connector instead of the parallel port connector, which shorts termination power to ground, which damages diode D800 and causes the PCB track to fuse open circuit.

Use Amiga Test Kit to run a complete memory test while you're at it, as defective memory can cause a lot of unexpected issues.

You will get problems if you mix fast memory DRAM types, i.e. if some parts are page mode and other parts are static column.  That can also cause issues such as system mostly booting if you disable CPU caches from Early Startup Menu, but will fail with the default of caches enabled.  More notes on A3000 memory here:
http://amiga.serveftp.net/A3000_HardwareGuide/memory.html