To be clear, the comment about null anaphora was more of a throwaway comment anticipating the objection that sometimes the topicalizing wa does mark the subject. While I have seen this explanation presented, and it is my prefered explanation, I would not necessarily call it pervasive. Now, for the explanation itself (unfourtantly, I am on vacation, so cannot check any of my references).

Japanese is a clear example of a pro-drop language, so using pronoun dropping (aka, null anaphora) as an explanation requires less justification than it would in English, where we only see it in specific contexts. Additionally, we see the topicalizing "wa" in various contexts, not all of which can be understood as subjects, so a unified explanation that can account for all of them would be preferred.

For example, consider the sentence

1) Mary-ga ringo-o tabeta

We can topicalize Mary with the following derivation:

2) Mary-wa Mary-ga ringo-o tabeta

3) Mary-wa anohito-ga ringo-o tabeta (Pro-form substitution)

4) Mary-wa ringo-o tabeta (Deletion)

Simmilarly, we can topicalaize ringo with

2) Ringo-wa Mary-ga ringo-o tabeta

3) Ringo-wa Mary-ga are-o tabeta (Pro-form substitution)

4) Ringo-wa Mary-ga tabeta (Deletion)

We also have the following sentence (kudamono = fruit)

Kudamono-wa Mary-ga ringo-o tabeta

Admittedly, I struggle to think of a context where the speaker would not drop Mary due to context, but that should not be relevent here, and I am sure that there exists better examples.

Notice that, under the null anaphora explanation, all three of these examples could be explained in the same way. If we were to explain the first example as wa being a subject marker, then we would need to explain the second example as wa being an object marker, and the third example as wa being just a topic marker.

I have seen an alternative explanation that describes topicalization in Japanese as a transformation rule. I have mostly seen this by researchers who view Japanese non-configurationally, who argue that a rule such as [ga/o] -> [wa] in a non configurational language is directly analogous to a movement rule in a configurational analysis. Even under this approach, you still need to account for sentences where the topic has no co-referential place in the rest of the sentence.

Further, even under this alternative explanation, I would still not call wa a topic marker. Rather, I would say that when the listener reconstructs the deep-structure, he uses pragmatics to infer what syntactic role the topic plays. Indeed, If you consider a sentences such as ringo wa tabeta and Mary wa tabeta you can see that there is no syntactic way to identify where the topic falls in the deep structure.