I haven't read through the original Unearthed Arcana cover to cover, although I've occasionally thought about picking it up if it's cheap next time I'm in Austin. There is a huge volume of material included within it, though, so I'll just give some bullet points.
- The addition of the "Comeliness" ability score seems more than a little silly to me. I get that Charisma isn't completely based on physical attractiveness, but adding a seventh score that seems more like a subset of another one isn't a good idea. And history seems to have agreed with me, as no other iteration of the game includes Comeliness as a factor.
(EDIT 2016-08-18: Having finally gotten the HackMaster 4th Edition rules, I see that they do in fact use the Comeliness score. Since a big part of that game is needlessly complicated rules and parodying the complexity and obtuseness of much of classic D&D, I think that my point still stands. Heck, even AD&D2's Skills & Powers book didn't use Comeliness as one of the two sub-scores for Charisma.)
- The barbarian class as written here is possibly an even worse party member than the assassin. An assassin could certainly, by the (Players Hand-)book, be played as extreme Lawful Evil, working with the party to further their own goals and realizing that the magic-user and the fighters are helping to watch their back when things get ugly. Not so for the UA barbarian; the rules flat-out state that barbarians will seek to destroy magic items whenever they are found. True, the barbarian was reworked as a more cooperative class for 3e and onward, and the AD&D2 barbarian "kit" isn't too bad, but the way it's presented here makes my brain hurt just thinking of the things that could go wrong.
- The thief-acrobat is interesting, although it seems like it could have been named a little bit better. I'd imagine that this awkward naming convention was responsible for the stereotype of RPGs having really long and convoluted class names (although B/X is not exempt from this; an 8th-level elf has the title of "Superhero-Necromancer").
- The cavalier is cool for those who want a game more focused on wilderness exploration, or a low-fantasy one modeled more heavily on medieval Europe. The rules for 0-level play are also good, presaging Dungeon Crawl Classics' "funnel" mechanic. Unfortunately, a lot of players were angered by the changing of the paladin (who is also modified for UA) to a subclass of the cavalier, rather than the fighter. This is extremely odd to modern ears, or those familiar with AD&D2; the arrangement of warrior/wizard/priest/rogue makes sense, so why add a fifth "group" that could easily be subsumed under the warrior group? This class also returned as a warrior "kit" in The Complete Fighter's Handbook for AD&D2, proving my point for me.
- Cantrips are a nice addition, to the point that I've considered adding them to my AD&D2 campaign. On the other hand, I like the default 2nd Edition idea of cantrip: an improvisational minor magic effect, that might add some wonder back into the arcane instead of making a spell "just another sort of laser pistol" (a quote from the book Authentic Thaumaturgy by Isaac Bonewits).