I've made some experiments, and found some rather surprising results.
First I cleaned out a Sony 8mb MC completely, and then copied onto it the folders /BEDATA-SYSTEM/, /BOOT/, /SMS/, and /SYS-CONF/. All those folders were taken from my main 32mb MC (Datel MAX Memory), which has worked fine previously with 'SMS exit', though it failed after the uLaunchELF v3.62 update. (NB: These tests all refer to starting mc0:/BOOT/BOOT.ELF from 'SMS exit' function.)
Now when I tested exactly the same setup, only freshly copied onto a clean 8mb card, everything worked perfectly...!!!
So, I erased all of those folders from the 32mb MC, and then copied them back to it from the 8mb MC, and tried again. But it still didn't work, although the symptoms did change slightly. (It now froze on black screen instead of freezing with SMS screen still visible, like it did earlier.) Still, at least some change occurred, so it was worth investigating further.
Next I made a complete mcPaste backup of the 32mb MC to HDD, so I could clean this card out completely as well,and then copied back the same four folders mentioned above. So here we have the same four folder setup I started with, back on the same MC where this setup consistently failed to work right earlier, but now it works perfectly there too...!!!
Apparently the launching functions have some problem either with the size of such memory cards (which I doubt), or with possible fragmentation of ELF files on it. I am inclined to believe in the latter, since many others too are having similar problems, and I think most of them are using standard MCs.
Fortunately these MC size or fragmentation issues don't seem to affect normal file operations (or I couldn't have copied my stuff back and forth safely), but apparently only affect ELF launching. It's still very strange, but not all that surprising, since we use a mixture of Sony and homebrew drivers, and Sony never intended ELFs to be run from MC in the way we are doing. So it's probable that they never debugged the ELF launching routines for coping with MC file fragmentation.
In any case, whether that conclusion is true or not, my experiment has shown one way of coping with these problems if they occur again. Though it's not very funny, having to backup/restore an entire memory card this way, it's still better than not having any way to fix the problem. Now it only remains to be seen if this method will fix it for others too...
Best regards: dlanor