The purpose of spanning tree is not to ensure that only a single copy of the frame arrives (although it does do that). It's purpose it to ensure a loop-free environment. There are some loop situations where multiple copies of frames are generated over an over. But there are also situations where the same single frame loops infinitely.
As for explorers, there can be many paths to a destination. Usually the first response from an all routes explorer is considered to be the most efficient and so that path is used. Since the path is predetermined, there is no need for a spanning tree, per se. However, IBM apparently came up with a proprietary spanning tree-like function. For example, setting a pair of parallel bridges in a "spanning tree" configuration means that only one of those two bridges will forward a spanning tree explorer frame. Either way, the source still needs to determine a path to the destination.
I don't know this for sure, but it seems implied that even a pair of bridges running the proprietary IBM spanning tree implementation would both forward an all routes explorer. Just seems to be intended to provide an option to reduce traffic generated by all routes. Sort of like the IOS, which can convert an all routes to a single route.
