What the binary does not settle on its own. The whole cartridge is explained byte by byte; this is what remains to be measured or decided.
DESCOMPRIME_PIEZAS (0x5496) builds 44 pieces at 0xE500 and the nineteen stretches at 0x5424 only point at 39. Three of them —41, 42 and 43— are junk read out of the code, and that much is clear. The other two, 0 and 1, are properly built from the dictionary and nobody names them. Whether they are a stretch that was dropped or simply where the numbering starts, the cartridge does not say.
In the openMSX measurement, out of 1112 demo frames exactly one went above 90 % of the available time (70,080 cycles out of 71,591). It has not been pinned down: it could be the screen change, which rebuilds the 192 characters, or a normal peak with a lot of sprites in play.
Everything measured —the cost of the name table dump, of the interrupt, the writes to the VDP— comes from letting the demo run, and the demo plays on its own without filling the screen with creatures. With ten of them and two shots in flight the frame will cost more, and a real game with a joystick in hand has not been measured.
The route has two halves of nine screens and the table at 0x7B62 chains them with two destinations each. On reaching the throne the map is turned over and the other half begins; what happens after the final screen, and whether the difficulty —which climbs to 0x20 and stops there— leaves the game playable, needs playing to find out.