So I have this crazy LUA script that writes a dtrace script to trace CAM I/O at the lowest level, then another bit of LUA to parse the results... Fun times that...
Most likely they were never converted to asciido.
Yea, there's two problems. geom is bogusly rejecting the fewer entries: that's explicitly allowed by the spec. geom should have a check for reserved space and issue a warning if it's too small (but it's not allowed to reject the GPT according to the spec). The other problem is mkimg isn't reserving enough space (the # of partitions is a red herring: that part is pedantically correct, but undesirable due to geom bugs). I have fixes for both.