Thank you for your help and informations.
I have tried three hard disks successively, all smaller than 4 Gb (200 Mb, 2 Gb and 540 Mb). The first one was in the Amiga when I bought it : the former owner was using it without a problem. I had never tested the second but it was given to me as working. The last one was taken from my old SCSI PC in which it works perfectly.
The three drives are terminated with their built-in jumper. I don't know if the first drive really is as the jumpers are not provided with a key, but with this drive the Amiga is the same setup the former owner was using without problem, so I guess this drive is terminated. The second and third drive are terminated without a doubt. Anyway the third drive was the only SCSI device in the PC as well. I don't have any independant terminator, but there wasn't any at the end of the internal chain when I bought the Amiga, so I assume the previous owner wasn't using any. To my knowledge he wasn't using any external SCSI device either, but I will ask him to confirm that.
The third drive (a Conner 540 Mb) is the only one that does not crash the computer before you launch the format command. However it crashes like the others at the end of the formatting at the earliest, at the latest when you browse through the drawers of a newly installed Workbench or try to read files you have copied. These crashes can take various forms, ranging from the "Software failure", "Out of memory error" and checksum errors I quoted, to screen freezes or scramblings.
The third drive bears a sticker with some physical informations such as the number of heads and cylinders : I notice that they don't correspond to what HDToolBox detects, although the latter does detect the right names and total size of the drive. I have tried to setup the drive again using the informations that are written on the drive, but some are missing : I tried to figure out the number of blocks per tracks and block per cylinders according to the informations I had, but don't know if they are accurate. And I didn't get better results.
Since the former owner was using the hard disk's built-in terminator, I doubt whether termination of the internal part of the chain is at fault.
I will test the memory with MemCheck tomorrow.
Thank you again !