First, the Knights / Development Card strategy seems too strong relative to Road Building (what you get along the way towards the largest army bonus has more intrinsic value than the extra development space you get when trying for the road building bonus).
Second, there is a lack of guaranteed growth due to high variance in production rolls, particularly during the early rounds of play.
The fixes (Tom Lehmann):
1. Reduce the largest army VP bonus from 2 VPs to 1 VP.
2. Rework the production roll mechanism. If the roll is either a 7 or matches a production site belonging to the active player, keep it. Otherwise, re-roll once, keeping whatever non-7 result occurs, rerolling again only if a 7 occurs (thus keeping the odds that a 7 is rolled to exactly 1/6).
Comments: You could try instead increasing the longest road bonus to 3 VPs and play to 11 VPs, but we're trying to find the smallest needed changes here.
Unlike various card and chit proposals to reduce or fix the variance in the production rolls, this proposal strikes at the real problem, the lack of guaranteed growth for all players in an exponential growth game. It does alter the probability distribution somewhat (reducing the odds that 2,3,4 and 10,11,12 are rolled since players tend to have porportionately fewer production sites on these numbers early on) but tends to the original 2D6 distribution as the game progresses and players expand their production sites to more numbers. (The amount of rerolling needed also reduces as the game progresses, which is nice.)
Note: These fixes are experimental and haven't been played a whole lot -- we may end up discarding them. Try them and let us know how they worked for you.