Backing up several steps: megadungeons are best used for exploration-style play. This is why they work well with open table campaigns and, somewhat paradoxically, in convention play. Both scenarios were used early and often in D&D's development, much more often than the continuous party format that arose after D&D became popular among adolescent players with relatively stable peer groups.
With a continuous party of 3 to 6 player characters used consistently throughout the life of a campaign, going into small "module-sized" dungeons that take 1-4 sessions to clear, having boss monsters is fine. When they're in the Upper Lowlands Dungeon of Death™ they are doing it to fight the King Zombie, not because the ULDoD™ is interesting in itself.
Megadungeons are different. On a given delve, a megadungeon needs to be able to accommodate players who have spent 50 sessions going into the ruins, and players who are only going in this once. Maybe they're going together down to level 5A. If that's the case, level 5A needs to be interesting as an exploration goal in itself, without regard to whether the PCs ever go down to level 6A.
That doesn't mean either that every room in your dungeon needs to have a full array of what's interesting about it, or that dungeon levels shouldn't tie together in any way; neither of those is interesting. But what it does mean is that every level and sub-level needs to be a goal in itself, that it's worth going into it, and to be interesting if the PCs go there. All of this breaks down if the sub-level is just leading up to a boss monster. If level 8 is just a lead-up to the boss at level 9, the players who are only there for level 8 are cheated. And that becomes increasingly true as you get into the low levels of Rappan Athuk.
More than that, the boss monster is antithetical to the "living dungeon" concept of a megadungeon. By definition, once you beat Orcus or the Elder Elemental God or whatever, the dungeon is done. Subsequent expeditions are never going to have the same gravity as the one that killed Orcus. That kills the multi-campaign potential of the megadungeon dead. After all, you're putting this much effort into designing a huge dungeon, it should be good for more than one set of adventurers. (And having the next group kill Mecha-Orcus is worse because it just opens up an arms race of increasing absurd power levels that the OSR is pretty good at avoiding.)
There's more to get into with the living dungeon idea. At its core it means you restock and redraw maps, but it should always reflect the influence the PCs have had in some way. This is why there is a "vision and re-vision" component to megadungeon design. Done properly, the megadungeon becomes archaeological itself, with cues and remnants from past campaigns in future ones, and a richer experience overall.
None of this means that there can't be intermediate goals within the megadungeon. You can create a faction boss so that everybody remembers the time they fought and killed the Red Witch on level 6 – but that's one among many parts to the megadungeon's lore. You can have puzzles and ideas that span four or five levels at a time so the PCs unlock the Vault of Artasius on level 8C and find the Warhammer of Magnificent Smiting. But the campaign could go on after that, and there can be more intermediate goals. The megadungeon will never be fully cleared and there will still be mysteries for future groups to explore.
If you're committed to an exploration-oriented game, it should always be possible that the PCs never kill the Red Witch or open the Vault of Artasius. And it should still be a place worth exploring, and the players should still come away with memorable stories. It should even be possible for the players to find half the puzzles for the Vault of Artasius, and solve them, and then go over to a totally different path in the dungeon and never finish it. The megadungeon from this angle is really a commitment to sandbox-style exploration, with the dungeon as the "walls" of the sandbox.
This standard, where each part of the dungeon is interesting enough for a drop-in player but the parts work together in a way that is rewarding for the long haul, is the central design goal of the megadungeon. It's a difficult note to strike, and one that I don't think can be managed while designing with a final boss fight in mind. Which is why I'd encourage a megadungeon to not have an end goal, even though there are many smaller goals within its structure.
