Strange Problem Amiga 2000

This might not be quite the right place, but the Amiga was made by 
Commodore after all...

Anyway, I have an Amiga 2000, PAL, the early model that uses a 
multilayer board and the A1000 chipset, that exhibits a strange problem.

The system is complete, including the memory card for the MMU/CPU-Slot 
which has been expanded to 1 MB and passes the memory test of the 
systest disk. I also own a another such card which also works.

There are 2 floppy drives, with DF0: having been replaced with a GOTEK 
floppy emulator using the latest FlashFloppy Firmware 0.9.29a.

The problem:

When you turn the Amiga on, it always starts and shows the screen 
prompting you to insert a bootable disk. Kickstart is 1.3 (got it this 
way). You can have it sit here as long as you want, makes no difference. 
If you then plug in a USB stick with a bootable image on it, the system 
will attempt to boot and, in most cases, crash with a GURU error of 
0000000B.00000000 or 0000000A.00C0xxxx. Taking the GOTEK out and putting 
DF0: back in, booting from a real disk only makes the crash less likely, 
it doesn't eliminate it.

If the system does not crash but boot, it will then happily run for 
hours on end doing whatever you ask it to do.

What I tried:

1) Suspected the PSU, replaced all caps on the secondary side. Measured 
the caps I took out (most were from FRAKO), all still good. Primary 
filter caps (made by Roederstein, ROE) were only checked, but are also 
still as good as new. PSU is made by 'ISMET', delivers up to 200W and 
shows no problem powering on with a dual filament lightbulb as a load 
for +5V and +12V. Load when on with the bulb is about 60W, more while 
it's still cold. New caps didn't solve the problem.

2) Suspected the KS 1.3 ROM and tried a different one. Didn't help.

And now it gets weird.

3) Decided to try a different CPU, Original one is a MC68000P8 from 
1987. Tried another MC68000P8, a MC68000P12 and a 68000 from Thompson. 
No dice. But when using a 68000 made in 1992 by ST, the crashes 
disappeared. Do I have 4 damaged CPUs? Seemed unlikely. All 68000 are 
the HMOS kind, none of them is CMOS. For grins, I tried an MC68010L10 as 
well. No change.

4) Decided to try one of my KS 1.2 ROMs. No more crashes with any of the 
CPUs I own. Tried a second KS 1.2 ROM. same result.

5) Tried the solutions from 3) and 4) multiple times to make sure I 
didn't just have a contact problem in one of the sockets. Always behaved 
as described above.


Anyone having an idea what might be the cause? The GOTEK delivers the 
data faster (meaning less latency, datarate is, of course, the same as 
with a floppy drive). Were there any changes to the trackdisk.device 
that between 1.2 and 1.3 that might explain this behaviour? And why does 
the one 68000 from ST solve it?

Since this A2000 will not be expanded but mostly used for games, I can 
live with KS1.2, but some part of me hates unexplained errors of the 
'all parts work by themselves, but not when used together' kind.

  Gerrit