I stress the focus of games, coherency in design and honesty with the player. While for the most part, these ideas get along well with arcade design, there are some tension points which I'm not quite sure how I feel about. This is an exploration written as a prescription, I'm still leaning towards high skill ceiling being better.
Huge depth and huge skill ceilings tend to be considered goals to strive for when you design games. But are they really? I don't think so.
Each game has a central focus, its thesis of sorts. The game logic is used to create, reinforce & future proof that thesis and coherency in design is used to clearly communicate it to the player.
While huge skill ceilings don't NECESSARILY undermine the thesis, they can and will eventually massively warp how your game is played and perceived. If this clashes with the core idea you want your game to represent, limiting a game's depth is desirable and even makes it better as a creative work, even at the expense of longterm player enjoyment (or maybe not!)
This warping can happen in very obvious ways, as the meta of your game develops it will usually stray pretty far away from your original vision - players will break mechanics, skip sections, invert risk vs reward, abuse and utterly fuck up what you thought to be the game's logic. More obviously, some games aren't meant to withstand pressure - if you want a casual game of luck, players optimizing it will turn it into an exercise in frustration.
The more subtle type of warping is psychological - the more players push the game, the more their personal demons will start to get mixed into the process of playing the game, eventually even making them resentful of the game. You see this happen a lot at high levels, players grow to kinda hate the games they used to enjoy. If your game is infinitely replayable, the only way to be done with it is to get fed up with it.
There are games which are simple, enjoyable and memorable because they do what they set out to do very well. I know everyone can think of games like that. And even though you probably won't be revisiting them too often because they lack depth and challenge, they will always linger in your mind as a very enjoyable experience. Rolling Thunder 2, while being a very simple game, is still one of the most enjoyable memorizers I have ever played, and made me realize why memorization is so valuable in games. The Shinobi games are in the same boat. The games are good because they are coherent and honest in their simplicity, and the idea that if they had a crazy skill ceiling I'd appreciate them more isn't necessarily true.
If hard skill ceilings are a bit too much for you - consider the value of soft ones while thinking about the central thesis of your game. I've thought about building in a hard-coded counterstop into a shmup, to very clearly tell the player to stop playing. 1 loop clears in shmups sort of fill this niche too. Even low-threshold S ranks do. Players gotta stop playing some time.