Thanks for your answer. It is somehow helpful, although I would have appreciated also some comments from people who actually tried to do what I mentioned with 2 separate probes and shared some practical tips/results. True differential probes cost more than the oscilloscope, an expense hard to justify with the rare usage I would make of them.

@Russell McMahon The 2 large filter caps in the primary section are indeed rated for 200V, I measured around 160V DC across them before the fault. The PSU is rated for 230VAC. The exact model is FSP250-60GTV. Unfortunately I could only find the data-sheet, but no schematic diagram. It is based on Power IC KA3511, like the one you linked to in danyk website. Interesting information, thanks.

Thanks for your clarification. I think I got it. Both scenarios will make the appliance-under-test float, but only the isolation transformer can ensure a physical break in the live electric conductor (via magnetic inductance) that will prevent completion of the circuit and therefore electrocution.

Thanks. Unfortunately your guess is wrong. As I wrote in my initial post, I did not short the probes at all. No distractions either, as I kept my eyes well focused on what I was doing, being aware of the danger. The bang happened the very moment the tip of the red probe touched the second pad in the image. I posted my questions because I feel I did not make any obvious mistake. I generally do not like wasting people's time with questions that can be answered by a DuckDuckGo search.

@nekomatic I do agree with what you write. And yes, I was holding the two probes with both hands (probing the two pads encircled in red), which is not safe at all. I had spent some time probing around the primary with the meter COM lead at the HOT ground (- of the large filter cap) with no problem at all. The moment I probed those two pads I witnessed the fireworks. I wonder if leaving the COM lead at the hot ground and probing those two pads separately could have prevented the fault.

I do understand what you write about the scope ground and I agree. I think it's dangerous to float a scope. A much safer approach is to make differential measurements. On the other hand, I still do not understand how an isolation transformer is any different from the earth-less lead I mentioned. You write: "Disconnecting the safety ground lead on the equipment under test doesn't remove the potential between the hot circuit and earth ground", but isn't this what happens with an isolation transformer? A fault to chassis will not be cleared as the earth connection is missing, in both scenarios.

Thanks for your answer. Let's suppose I had connected the ATX PSU to an isolation transformer. Am I right to say that it would not have protected the PSU from the over-current inrush? Its only aim is to float the equipment under test to prevent electrocution in case I came in contact with a live conductor (which did not happen). And talking about safety, am I right to say that an RCD only (with no isolation transformer) wouldn't have helped (tripped) either if the over-current fault did not cause a current leakage to ground?

I posted this request for help because, to me, the mistake I made is not obvious. From some of the judgemental comments I read above my mistake seems to have dared to probe a high voltage area. While I have read books and many interesting posts in this forum about power supplies, how am I supposed to learn if I do not take the courage to experiment? Far from being reckless, I think that the only way to learn is through direct experience.

@Russell McMahon. The meter was on high range volts DC and it's still working fine (as very little current is drawn owing to the high input impedance, if I am not mistaken). It's a Brymen BM257 (1kV cat II, 300V cat IV). I'm in the UK, so mains is 230VAC. While I am no expert, I am very conscious of the beginner's fatal mistakes when using a DMM to probe mains. What do you mean by "orientation of the parts"?