Phrasal complexity is more than just examining nominal group structure. It quantifies nominal group complexity in some way (e.g., the depth of structure (either in terms of constituency or dependency relations), the total number of connections, etc.
For me, the hardest part is coming up with a complexity metric that actually makes sense. This is actually an area where psycho linguistics has lots to say, in terms of the readability of a given nominal structure, using eye tracking, etc.
I haven't seen any work within SFL on this, mostly by post-chomskyans and cognitive linguists.
In terms of SFL theory, this relates to the type of complexity that is typical of written mode: lexical density, and, as Halliday & Matthiessen (2014: 726-9) point out, the nominal group is the primary resource for increasing lexical density.
Lexical density is quantified at clause rank — rather than group rank — by dividing the number of lexical items by the number of ranking clauses.
Moreover, in terms of semantic complexity, lexical density involves grammatical metaphor, a junctional construct (Halliday & Matthiessen 1999: 46, 272), embodying the meanings of both the metaphorical and congruent grammatical realisations.