I very much like the idea of individual weapon proficiencies, reflecting the in-game notion that a fighter - while vaguely handy with any weapon hence their lower non-prof penalty - has his-her own particular weapon preferences, and those are what he-she has chosen to focus on.
I don't like it, and more importantly I don't think there's anything realistic about the kind of narrow weapon proficiencies that 1e and 2e had. People who were competent, experienced melee fighters didn't need large amounts of training to switch from a sword to a spear to an axe, or worse from a sword to a slightly smaller sword or from a poleaxe to a poleaxe with a spike on the back. I mean, can anyone find a historical source that troops switching from one style of polearm to another needed months or years of training to learn the new one?
Tiny numbers of weapon proficiencies just isn't realistic. If it floats your boat go for it, but don't do it in the name of realism.
The point of small-weapon initiative bonus is not to reflect who will usually strike harder or inflict more damage, but to reflect who will usually strike faster. Assuming vaguely-equal levels of skill, the person with the dagger will be able to pull of a couple of moves during the time it takes the sword person to wind up and swing...particularly if the dagger person can get inside the swordsperson's guard.
Right, the fact that initiative is about who will strike faster is why the sword should have a HUGE initiative bonus compared to a dagger. The quarter-second of 'wind up' isn't significant, what is significant is that the sword wielder can hit effectively while the dagger user isn't even in range. The idea that someone holding a dagger can get off multiple moves in the fraction of a second it takes for 'wind up' with a sword simply doesn't match reality - the dagger user has to wait for the sword user to attack first to have any real chance of hitting without getting killed himself.
In sparring and fencing situations the dagger fares somewhat better than it would in real life with full-strength swings and a person not concerned about the dagger-wielder's safety, you can certainly find videos of dagger vs sword where dagger wins. But the more realistic they make the situation (using full speed blows, for example), the harder of a time the dagger user has. And even in cases where you watch video of someone using daggers to win against someone with a sword, you will almost NEVER see the dagger user able to strike first - successful dagger users wait for the sword user to swing, then exploit the sword user being off balance to attack. If you always have to wait for your opponent to swing before you do anything, you certainly do NOT have the initiative in the fight.
I don't have a problem with letting dagger wielders be effective melee combatants in the game, but it's not actually realistic, and giving them an initiative bonus is anti-realistic, since in reality they will almost never strike first.
The chance of messing up, when pushing the edge of one's skill, doesn't change much as the skill improves.
In real life, the chance of seriously injuring yourself goes WAY down with increased skill; stabbing yourself in the leg was incredibly rare on the battlefield. In game terms, most critical fail systems have the chance of seriously injuring yourself go WAY up with increased skill. A 1st level fighter might get a critical fumble (roll a 1) in 1 of 5 fights, a 20th level fighter will typically get multiple critical fumbles per fight because he'll be doing more than 20 attack rolls in the same time span.
As an example - I'm not much of a typist by any means but I'm almost certainly better at it than I was 20 years ago, at least in terms of typing faster. My number of mistakes per line typed, however, probably hasn't changed much if any over that timespan - I just make those mistakes more frequently in clock time because I'm typing faster than I used to.
Mistakes per line types is not a good comparison, those are more akin to a regular 'miss'. How often do you injure your finger or knock the keyboard off the desk while typing? Critical failures are just not a routine part of a highly skilled task, but critical failure tables make them into one.