Sorry for the long delay.
Discussion on this proposal was generally in favor of it, but several people expressed reservations about increasing the apparent complexity of the library's lazy subsystem. The Core Team agrees that it's unfortunate that decisions like this are exposed to clients of the library; this objection might be addressed by a feature like opaque result types. The Core Team also observes that the performance benefits of this change currently appear to be negligible because of improvements to the optimizer. Accordingly, this proposal has been rejected. If future developments make it profitable, we can reconsider it then.