My test for (full) compositionality is: Given the components (in this instance, cinco and ciento) plus any rules of inflection and phonological adjustment (e.g. ciento -> cientos), can a speaker reliably produce the correct form? That's not the case with quinientos. Given that word, you can work out the relationship to quinto and such, but you'd never guess that form existed based on the information above and the examples of un ciento (not *uniento), doscientos, trescientos, etc.