Why does Wheeler's bag of gold solution contradict the holographic principle and a black hole does not?
The basic idea of the bags of gold paradox is that naively there are too many degrees of freedom in the bulk. For slicings which have a very large volume (e.g. exponentially increasing volumes as in FRW universes), one can naively place many scalar excitations in that volume, and consequently the entropy counting (which goes as volume) can increase the Bekenstein-Hawking entropy of the black hole. Hence the paradox.
We will work in AdS as there we best understand quantum gravity in this setup. As you mentioned, one can construct spacelike slicings in the case of AdS black holes which can have increasing volumes. See the worked out case for "maximal volume slicings" in Appendix 1 of this paper. Here the paradox is this, these maximal volume slicings have increasing interior volume with the Kruskal time, and any scalar field will naively have a similar overcounting problem as in Point 1 above.
Consequently there is an overcounting problem in the AdS bulk, which arises either because we are treating all the volume excitations independently, or because there is no interior of black holes in quantum gravity and therefore no map from the interior of the black hole to the holographically dual CFT (as the firewall papers suggest). I strongly think that the map exists from the interior to the CFT (this is a possible state-dependent map), and therefore the bags of gold is a question why the degrees of freedom are overcounted here.
Note that the holographically dual CFT description doesn't have any paradox. This is because conventional counting of the CFT degrees of freedom have precisely given us the Bekenstein Hawking entropy in numerous cases. The CFT doesn't know about which spacelike slicing we are working on and gives the correct BH entropy.
EDIT: It might be helpful to look at this recent paper which studies the bags of gold in AdS, and proposes a resolution.