I don't think there's any particular reason that being updateless should be useful in general - it's just that "find the winning strategy, and then do that" is equivalent to being updateless, and works really well on most problems.

In fact, I'd expect a computation-aware decision theory to not say "find the winning strategy, and then do that," because it couldn't always expect to do that except in the limit of infinite computing time (even if the problem is guaranteed solvable) - and that limit would also eliminate logical uncertainty.