Many of the Cursers are so strong that they're must buys. Maybe give them all a clause so they only work while the Curses have not run out: Every other player gains a Curse, if at least one player does, do your bonus thing.
The bonuses is so minor anyway. Witches become Moats without the reaction effect -- even less than a $2-cost power. Familiar becomes a cantrip that does nothing at all. Sea Hag becomes a dead card already. Mountebank retains some power after the curses run out, but by that time the curses that are already out there will defend against it, so it sort of nerfs itself.
I really don't see what you're going for here. Or, for that matter, what problem you're trying to solve. Must-buy cards aren't inherently a problem.
Well, I think any "must-buy" is sort of a problem, because it takes some strategy out of the game. It's very hard to win if you're not actively competing in the Curse race, especially when playing multiplayer. I'd like to see some good examples of games in which one player buys a Curser, the other doesn't and the non-Curser wins. This may cure my addiction to Cursers.
Sure, most of the Cursers don't do much after the Curses have run out, but then the damage has already been done. You have 10 cards that clog up your deck, let alone the -10 VP that you have to fix.
Councilroom lists these stats (Card - Win rate with - Win rate without):
Mountebank - 1.09 ± 0.00 - 0.82 ± 0.01
Witch - 1.08 ±0.00 - 0.85 ± 0.01
Familiar - 1.05± 0.01 - 0.90 ± 0.01
Sea Hag - 1.04 ± 0.01 - 0.94 ±0.01
Disregarding Prizes, VP Cards and Treasures, these are all in the top 10 highest win rate with.
If we sort on win rate without, only Sea Hag falls out of the top 10.
How to nerf these cards? Well, I have always thought it would be nice to have some Curses that are also regular cards, like a card that has -2 VP, but also maybe +1 Card / +1 Action (or just +1 Card if it's too strong) that players can choose to take. Of course, when Swindled, the Swindling player may still choose regular Curses for the others to receive.