Raz offers a different argument, in his 'Reasons: Practical and Adaptive', p.7:
because there is no possibility that the lesser reason for belief serves a concern which is not served better by the better reason there is no possibility of preferring to follow what one takes to be the lesser reason rather than the better one. The possibility of akrasia depends on the fact that belief that a practical reason is defeated by a better conflicting reason is consistent with belief that is serves a concern which the better reasons does not, and which can motivate one to follow it.
The thought here is that in practical cases, there is at least something to be said for acting on the lesser reason. If I can save either one life or two others, there's something appealing about the former option, even if the latter is better on balance. The former reason is outweighed, but not entirely defeated. Epistemic reasons, in contrast, seem to be defeated and not merely outweighed. As Raz puts it (p.6):
The weaker reasons are just less reliable guides to one and the same end [viz., truth]. There is no loss in dismissing a less reliable clue.
Or, in other words, practical reasons are pro tanto reasons, exerting some degree of force, whereas epistemic reasons are merely prima facie reasons -- liable on further examination to turn out to be no real reason at all.
But I wonder whether it is true that "there is no possibility that the lesser reason for belief serves a concern which is not served better by the better reason". Sure, epistemic reasons serve truth, but practical reasons may be thought to serve a monistic good (e.g. human welfare), and that doesn't prevent tradeoffs between different instances of this end. So let's consider a case where there are competing truths (analogous to the competing lives in our earlier example):
Suppose I am assessing two competing belief-sets or comprehensive theories/ideologies (T1 and T2) regarding some area of discourse D. Though both seem flawed, I am aware of no alternatives which are more coherent and plausible than these two, and can see no promising way to combine them. Further suppose that T1 seems on balance the better theory -- more likely to be true, or true in more of the sub-areas of D that matter. However, I nonetheless think that T2 is much more plausible regarding some specific sub-area Ds. I am so drawn to T2's account of Ds, that I end up accepting (believing?) T2 overall, even though I judge that I have more reason to believe T1.
Does that sound like a genuine case of epistemic akrasia?