catchException
  :: forall a e
   . (Error -> Eff e a)
  -> Eff (err :: EXCEPTION | e) a
  -> Eff e a

 would be unwieldy to express using typeclasses, requiring at least three constraints. I also find this style easier to read than constraints, as it requires no mental substitution. E.g. if I see 

(Foo m, Bar m, Bar n) => Baz -> <stuff ...> -> m a

When I get to the end of the type, I have to go back to the beginning to figure out what m is. I can't read left-to-right. This happens a lot with constraint-based monad composition. 

Another advantage is that the Purescipt example uses a concrete type, which is often easier to reason about than "ad-hoc" typeclass abstractions like MonadRandom. However, it looks like you still get the flexibility of ad-hoc typeclasses, because you get to pick any function that discharges the effect type in the given effect monad. 

Like I said, I have not used it, but these are what I've noticed from topical observation. 

Apologies for the formatting; copying that code example appears to have confused the iOS mail app. 

Cheers,

Will