This rings really true to me. I only had played the shareware version of quake and only recently played the full game, and the second half really felt like padding.
Some gimmicks in particular were really overused later on (I specifically recall that at a certain point I could just tell "There's going to be an ogre hidden in this corner").
I'm not sure if there was also some other reasons - maybe they decided to use some of the best designs for the shareware, maybe ran out of space for more enemies or maybe there's just way more nostalgia for the first levels - but I'm happy to see that some people share a similar opinion.
Though it certainly is nicer on my wallet to normally grab games on sale.
Also, the distinct style of the last episode is easily explained by the fact that all of its maps were built by Sandy Petersen. E3 was mostly American McGee, while John Romero and Tim Willits worked on E1 and E2.
I'm not a fan of Petersen's maps either, but they are regarded (and liked) as quite unique, compared to the rest of the game.
I had a pirate pre-release copy of Quake that I'd wished I'd saved. But in any case I do remember there were changes to the maps all over the game, so the levels were definitely not done in order. The biggest difference I remember is the ending in the final game is totally different. The pre-release had a more Doom style waves of monsters fight on a sort of giant terrace. I didn't particularly like that doom style ending, but also felt the release game's ending was kind of an anti climactic gimmick.
The better single player levels in Quake are actually really intricate and well-designed single player levels too, in exactly the way that Doom levels were great.
Some of the enemy design in Quake is actually pretty good, too, with sharply distinguished silhouettes between enemies, and good discrete gameplay property differentiation between them - well, at least for the zombies, the fiends, and the shamblers. Some of the others are fine, too (scraggs, grunts).
I think the bigger issue is they bit off way more than they could chew.
Specifically, Doom's single player mechanics thrive on having hordes of enemies interacting with highly interconnected levels in interesting ways, making space management a big part of the single player game play. Trying to figure out where you're safe to pick a fight, or how to steer the hordes around to create a space where it's safe to fight, is a big part of the draw. But, because rendering 3d monsters was so expensive compared to 2d sprites, Quake couldn't do hordes when it shipped, and fighting just a few enemies at a time (with their health highly jacked up) was a totally different experience. It could have been made to work, but they would have had to stray a lot further from the Doom gameplay recipe than they did.
And they also would have needed, I think, a lot more monster variety. A single player game with 30 different enemy types that were interesting and differentiated would have been much stronger.
That's my two cents, anyway.
IIRC (this is well-documented if you want to double check), Tim Willits made most of the Episode 1 maps, John Romero made most of the episode 2 maps, American McGee made most of the Episode 3 maps, Sandy Peterson made most of the episode 4 maps, and John Romero made most of the level 1, military base themed maps in each episode.
The episode 4 maps are often barren, lacking in details, and missing much of the beautiful interconnections of earlier maps... but this is also true of Peterson's maps from Doom (he did a lot of episode 3 in the original Doom, IIRC). So I think it's more of "this guy might be a strong game designer in a lot of other contexts, but the specific needs of making cutting edge Doom/Quake style maps isn't a great fit for him".
I was at Raven Software at the transition from the Doom engine and other 2.5D engines to Quake (and then Quake 2, and then Quake 3, and then Doom 3), and there were a number of existing designers who were fine game designers in earlier, 2d contexts who found their skills severely out of sync with the changing demands of 3d map making, and most of them eventually had to transition to other roles or leave the industry.
His Quake maps especially give me the impression that he was more into trying to come up with ideas on what is possible for the player to do in the freedom allowed in 3D space than how to make a good looking environment (especially in Quake's theme that didn't really have to conform to any realistic constrains and could have shapes floating in space, physically impossible architectures or whatever).
One of them is the nature of interactivity in the levels themselves. There's a spectrum between having a game grammar made of distinct discrete interactive reusable objects and then building unique situations by assembling them in interesting ways, versus having (essentially) unique scripted traps or interactive things or set pieces that only show up in one place. Older action games that inspired Doom tend to draw from that former tradition; a lot of FPS games that came after Quake tended to go more down that second road. Quake's trigger system specifically opened up the door to a rudimentary kind of visual scripting that made the latter style of design more possible in a way that wasn't possible in Doom (although it was possible in Hexen via HexenC(?)). I think you could say that that style of design really came more into its own with Half-Life, which foregrounded unique interactivity grounded in very specific, themed levels much more clearly. Doom at its best seems like it's much more in the design space of, say, Robotron and old Mario games. Fewer unique set pieces, much more focus on discrete interactive toys to be recombined... and given id's background with Commander Keen and their earlier recreation of the first level of Mario 3, this design influence shouldn't be a surprise. Anyway, Quake feels like it is at the intersection of these two styles of design.
I think it is true that Peterson did try to go more down that second road of design in the episode 4 maps in a way that there was less of in other maps, and that it interesting.
But the other thing that sticks out to me more so, in terms of level design, is about the way the space is shaped. A lot of the very best Doom and Quake levels have a tendency of having different parts of levels intersect and interact in interesting, playful ways. The order that you see areas is different from the order that you hear areas is different from the order that you can attack into or interact with areas is different from the order you can move through areas is different from the order that different kinds of enemies can move through areas or attack areas. And that changes as you progress through a level, get keys, and activate switches. There's a tendency for levels to start somewhat linear and movement constrained but give information about later areas in a somewhat more non-linear, tantalizing way, and then as a player progresses, for the player's movement in a level to become more like a multiply connected graph as switches, keys, and activated lifts make a lot of one-way paths become two-way. And that style of design plays to the strengths of Doom and Quake using BSPs for levels as their fundamental data structure - BSPs specifically make these kinds of weird and surprising visual and physical intersections between areas manageable in terms of computational performance on 90's era hardware.
Whether or not someone considers the design approaches I just outlined appealing is fundamentally an aesthetic issue, obviously - there's no one right way to enjoy a game. But my general sense is that the Sandy Peterson maps in Doom and Quake tend to explore these approaches to play much less than the maps made by other designers.
afaik Romero was fired for not working. Cubicle walls came down because Romero was playing Doom on the clock instead of making the game(his form of protest for not making an RPG or something). If you look at Quake code most broken/lazy stuff was done by him. Example https://www.youtube.com/watch?v=DEkjDkr0Qmc
The only real complaint I have is the use of the Spawn enemy, which is generally considered the worst designed enemy in the game. But the enemy placement in episode 4 has the advantage that it makes relatively little use of the Ogre, which I consider the second worst enemy, because it has too much HP for something so common, and is too predictable on Nightmare difficulty (to the point that some people say Hard difficulty is actually more difficult than Nightmare).
That's most indie game devs of the period, but more so at iD Software which splintered the team during Quake's development due to internal squabbles, causing John Romero and others to leave iD.
You can even tell that from the quality of Quake's levels which start with beautifully crafted and intricate levels, and as you approach the end, progress into "whatever, let's just ship it, this is gonna sell" kind of levels that were mostly just filler to pad the play time.