A trap that many ARPG designs fall into is overspecialization in the face of relatively permanent character advancement decisions. Path of Exile is a great illustration of the pitfalls of optimal play revolving around specializing your character using permanent decisions over a long span of play time.
PoE has an extremely large and intricate skill tree which allows significant specialization: down to the weapon type level--swords, axes, etc and element (fire, lightning, etc) level. You'll need to look for specific weapons and items that suit your specialization choices.
This is damaging to the ARPG model. ARPGs are a combination of various treadmills that complement one another: loot, character level, ability levels, and perhaps a few more (Din's Curse had "reputation" which would grant you a relatively viable rare weapon upon leveling up). An emphasis on specialization limits the character construction decisions (once you've picked the specialization, you jump pump points into that indefinitely) and loot decisions, cutting off significant branches of the decision tree. The more equipment-related specialization, the fewer pieces of loot a character will find that can possibly be upgrades. A character can easily go from having a 20% chance of finding a relevant piece of loot because he specializes in one-handed weapons, to a less than 5% chance because he has been forced to specialize further, perhaps into one-handed swords, in order to continue getting damage and/or accuracy bonuses from new passive skills.
ARPG players are familiar and fond of systems that involve an extreme amount of character choice permanence. Diablo II's attribute and skill points were non-refundable, and the pattern of severely punishing the uninformed by making character decisions irreversible does not enter the consideration of the majority of ARPG-players who grew up playing Diablo and Diablo II. Titan Quest and Din's Curse made a step forward in this regard by allowing point-by-point respecs at escalating prices. Torchlight, the most casual-friendly ARPG I have played, doesn't have respecs built into the game at all--the only way to respec is to use a console command to spawn a respec potion.
Overspecialization leads to minute-by-minute play being less interesting. If you need to invest a high percentage of your points into a very small number of abilities to make them effective, then you're naturally going to be filtering your ability choices by which abilities you have spent points on. Abilities available are further filtered by what loot you have available to you. In the majority of ARPGs you end up in a situation where you have at most two or three useful abilities--sometimes you'll only have one. When you have no more than two or three tools to work with, it's harder for combat to remain engaging.
Spamming a single ability and watching everything die can only remain fun for a little while unless you are specifically looking for a relaxing experience. Unfortunately, the theorycrafting involved in making an adequate character often is beyond the interests of a relaxed player, so they wouldn't get to the point where they could effectively do one ability spam. The mechanical systems thus lead to patterns of play which are not appealing to the kinds of players that have the capacity to use that style of play.
Permanent point-investment schemes also lead to perverse incentives for building characters. You really shouldn't spend points on skills beyond what you specifically need to get to that highest level skill that you want to specialize in. When optimal play is to not participate in the character advancement and be underpowered for tens of hours so that you can be somewhat above average later, the game clearly suffers.
The reasoning and evidence above would indicate that specialization-focused design is poor design.
So why do players seem to like it so much? A game like Diablo II is a complexly layered system of rewards that vary in intensity and frequency in such a way as to draw us in and addict us. Players have trouble separating the fun and not fun mechanics of a game when they are layered as they are in Diablo II. The rewards systems are strong and interleaved into all other mechanics of the game, so it's hard to separate skill tree manipulation and loot sifting from the enjoyable feeling of character progress. Those activities do contribute to character progress, but their design isn't trivial to separate from that positive feeling.
Good ARPG design is much more than simply causing the player to have the positive feeling of character progress, it's optimizing that feeling to happen in as intense and frequent a way as possible without it being diluted by overexposure. It's a difficult balancing and timing act, and different players have higher engagement at different points along the spectrum between constant rewards and rare rewards, large rewards and small rewards.
In games that feature overspecialization, you'll notice that combat tends to reduce to the repeated use of a couple of skills at most. This trivializes combat and turns it into a chore. When combat is a chore, in order to enjoy the game you must enjoy the minmaxing of character construction, which in ARPGs is done via loot sifting and planning/spending skill and attribute points. So the games' fanbases naturally require their members to enjoy that minmaxing and not mind somewhat boring combat.
ARPGs can have exciting combat and enjoyable character advancement. Diablo III is proof. I hope to discuss Diablo III's success in a future article.