Being able to abstract over the monad leads to monad transformers of the `transformers`, `mtl`, and `extensible-effects` varieties. You can argue that those aren't valuable, but I don't think you can argue that type-checked generic monads are critical for achieving them.
In practice, 75% of the time that I use monads I am using some kind of monad stack. I'm sure you can argue that this is unnecessary or that I only do so because Haskell affords it... but on the other hand, I believe that transformer stacks do a good job of elaborating the space of effect types with an efficient, expressive language. Without them you either elide side effect reification or build a lot of one-off monad stacks.
In practice, 75% of the time that I use monads I am using some kind of monad stack. I'm sure you can argue that this is unnecessary or that I only do so because Haskell affords it... but on the other hand, I believe that transformer stacks do a good job of elaborating the space of effect types with an efficient, expressive language. Without them you either elide side effect reification or build a lot of one-off monad stacks.