It's fun to hear the diskspace argumentation in 2023....if you decided to collect MAME (and maybe software lists) you shouldn't care about the few extra size of such "clones of clones" at all....each upcoming laserdisk chd will be more than all the files together...
I personally think "merged" or "split" was never a question of "hey I can save a few MB"...it's more about how you look at your collection. The "full merged" guys tend to "I want only one pacman archive and not pacman1 to pacman 20", while the "split merged" guys are more like "MAME support 1003 sets, I have 1003 archives and I can see the difference of pacman1 compared to pacman2...
"This way it does save (some marginal) space, doesn't it? as in my pit fighter example, rev3 becomes a parent for rev4" ....well, yes, but "becomes parent" is something romvault made up. There is no indicator in MAME's -listxml file for it. And yes, it saves marginal space
If you won't use subfolders for full merged sets, then you a) won't have the pitfighter roms doubled and b) you would even save more space since the folder name is not in the archive....
"What about split then? will CMP also include the rev3 roms into the rev4 zip?"
Split merged: you have 1 archive for each set, parent archives only hold the files for the parent, clone archives only hold files which are not marked as "merged". There is no sharing between clones
Full merged: you have only archives for parent sets, where the archive also holds the files for each clone.
To answer your question:
rev3 would hold: 136081-3028.05d, 136081-3029.05b, 136081-3030.15d, 136081-3031.15b (but not 136081-4028.05d or 136081-4029.05b)
rev4 would hold: 136081-4028.05d, 136081-4029.05b, 136081-3030.15d, 136081-3031.15b (but not 136081-3028.05d or 136081-3029.05b)
"merged is just the same as split, but clones are just included as subfolder in the parent/main rom, which could save some space because of zipfile header overhead"
the subfolders part is not really a part of merged sets. Originally there were no subfolders (and MAME does not need subfolders), subfolders were introduced first where it made sense (e.g. if you have clone files with identical name but a different hash)....but surely they give a better idea which file belongs to which clone if you always use them....but that latter argument also is an argument to have the clones-of-clones in each subfolder ;-)