[Background: I'm writing this as a developer whose firmware's in flight on several substantial satellite missions. I've developed attitude control systems, working directly with RW hardware engineers.]
As Hash says, there is a lubrication distribution issue with conventional mechanical-bearing reaction wheels. The result is increased wear, leading to wear particulates that become embedded into the lubricant, further accelerating the wear process.
There is a solution to this: magnetic-bearing RWs. These have no physical contact between the rotor and the static assembly, so have a mechanically indefinite lifetime. They also have the benefit of greater torque, and less micro-vibration. Unfortunately, these are both much more expensive, and can be rather heavier, than mechanical designs.
If I may expand the scope of the question slightly (but remain on topic): the situation is compounded by the finance model for space missions. In order to be able to declare a mission success (and hence funding for future projects), mission designers set a very low bar for success. Many, perhaps most, missions are designed with an unspoken expectation that they will exceed the success threshold by a very substantial margin. The problem is that the mission manifest will be based upon the bill of materials to achieve that low success threshold, something that will be under heavy scrutiny from budget-conscious managers and/or customers.
Thus, reaction wheels will be chosen that satisfy the base mission requirements, but are not necessarily fit for purpose for a substantially extended mission. Unsurprisingly, we then see early failures during the extended missions.