The ones that uses 3.3V is usually SMD and much smaller and is not the ones you want.

The ones that are 5V are through hole and in a metal can in two types square and rectangle. Yet there is a black plastic package with four J leads still is 5V that is usually used on yours is 5V also.
You can use either metal or plastic as described using 5V but not the tiny ones as they are 3.3V and not compatible with yours.

Cheers,

I've so far been using the metal rectangle cans. They seem to work at 5V, but the output is not universally the same (as it seems), which could be the cause of them not working. I could not find anything about this online. Did you run into this?

I've played musical crystals on IBM machines either side of that one, the earlier one, there's a BIOS lock based on a timing routine that stops you using a faster CPU/oscillator... the later one, there's about 4 different jumpers to switch over to enable 25-33Mhz switch... So I'm saying it might not be that your crystals are inadequate, but that you have IBM problems... deliberate prevention of you doing it, or obscurity of the settings data that lets you do it.

Ah, interesting! I checked it out right away. On my 'planar' there is a jumper (JMP3) that is 'CPU speed select' according to https://ardent-tool.com/76_77/Lacuna.html , but without any guidance as to what is does. I tried all possible settings (left, right and completely disconnected) with 60 and 66MHz crystals, but no cigar.
An argument against it is the section on the same page that discusses overclocking the planar to 40MHz (80MHz xtal), which for some works fine and for some results in issues with the video card, but no mention of complete non-boot.

One thing I want to try is to use an old 40MHz crystal I have left from upgrading a different PS/2. Although going in the wrong direction, if that works it's more likely the other crystals have some issue (maybe voltage). Man, if only I had a signal generator in this range....

Alright, I soldered some legs on the 40MHz crystal and popped it in. The machine boots, but gives error 10200. After a run through the system diskette, it does boot to windows 95, so seems to work stable with a 20MHz bus, instead of the stock 25MHz. So, I can go lower, however its not completely happy about it.
I also did some oscilloscope measurements on the crystals I have, while in the machine:

40MHz: 10Vpkpk - system works, but boots with error 10200
50MHz: 4Vpkpk - systems works normal (=stock speed)
60MHz: 4.45Vpkpk - nothing on power on
66MHz: 1.8Vpkpk - nothing on power on

So what to make of it? The 66MHz might be too low voltage, so lets forget that one. The 60MHz however should conceivably work, but doesn't. Is this the IBM trying to be clever and lock me out due to an unexpected frequency, like mentioned above? But what about the text on overclocking found on Ohland's website, where a 80MHz crystal seems to at least somewhat work in a multitude of machines? Any suggestions are welcome!

Further insight https://www.ardent-tool.com/PS55/Overclocking.html

Thanks a lot! So, at least for the model 55, it seems you can get about 10% out of whack upwards before tripping errors. The solution is to either modify the bios, or apparently swap crystals after POST. Maybe I could cook something up for the latter...

What about a different clock gen?

New clock gen for tyan s1564d (Research)

Some interesting info in this thread. At one point it was suggested that a custom clock gen could be made. Ultimately we haven’t had any success though

Little follow up. My dad brought his signal generator for some experimenting. Only problem was that it only goes to 60MHz, which is not really the 66MHz I needed, but close.
We hooked it up to the crystal socket, initially at 50MHz (sine between 0 and 5 volt, so bias of 2.5V), at which the IBM booted normally. At 60MHz it didn't show anything. Going down, we determined that the maximum frequency for a boot was 53MHz.

Next, we tried changing the frequency on the fly. This is not as smooth is it may seem, changing once the OS is booted (DOS or win95) freezes the system. Also changing during the POST results in a freeze. The only moment it works is just when the BIOS wants to start counting the memory, then we would go from 50 to 60MHz. So it definitely seems there is some sort of built in protection that stops the computer from booting if the frequency is out of bounds. There is a jumper on the motherboard which is supposed to be ' frequency select' (JMP3), but that didn't do anything (also not with the 66MHz crystal I tried earlier).

But even changing on the fly at the exact right moment wasn't great; win95 would start displaying kind of 'static lines' during the startup screen and freeze. Probably due to some issues with the SiS graphics. Booting into DOS however worked, and was stable. The processor was now running at 120 instead of 100MHz and would even run a DOOM benchmark, going from 3084 to 2710 points, so a real increase, but not earth shattering of course.

In the end, the fact that windows wouldn't boot and the extremely hacky way to change the frequency at the exact right moment during startup we not really encouraging, so I gave up and soldered the original 50MHz crystal back in. This thing clearly does not like to be oc'ed.

So need to find a 53mhz crystal?

Haha 😀 Nah, I'm a bit worried its not really worth the effort.