The underlying problem here appears to be that the TRANSUM post-dissector is not only adding stuff to the protocol tree, which obviously doesn't need to be done if there is no protocol tree, but is also doing *analysis* of the packet information.
The latter of those should *not* be done in a dissector - it should be done in a tap.
Unfortunately, *currently*, taps are run after all dissectors, including post-dissectors, are run, which might not work for this purpose. If so, what we probably would need here is to have a mechanism to allow taps to be run "early". For now, we could define "early" as "before the post-dissectors are run".
