(As a side note, Wikipedia's toy example of "STV in a party system" considers a case where three parties field candidates for 5 seats and receive 48%, 45%, and 7% of the vote, resulting in those parties respectively winning 2, 2, and 1 seats. This is about as bad in terms of proportionality as any other system [Party A has 686% of Party C's votes and 200% of the representation; Party B has merely 643% and 200% of the representation], which isn't a surprise; 5 seats don't allow for proportionality unless you're lucky enough to see a 63-37 split in the electorate. But if party-based proportionality is a concern, the correct answer is fairly obvious - the three parties should receive 2, 2, and 0 seats, with one seat going unfilled. It is interesting to me that no one ever seems to contemplate this as a possibility.)
I don't think that the approval voting scenario you suggest is good though. If I want to elect five candidates, I want 5 candidates that represent the diversity of the population. But it seems to me that your suggested method will elect 5 candidates of the largest voting bloc, not at all what we want.
The example is very explicit that votes are cast for parties and all of the benefit of receiving a vote goes to the party for which it is cast. As such, the model is only coherent if there is no such thing as voting for a candidate.
But really, what exactly the original example intended to convey is a pointless pedantic argument I'm not getting into.
The crucial point is that you absolutely can have multi-member proportional districts where you vote for specific candidates. So to someone who advocated for a such multi-member districts, the contention that you have to give up voting for individual candidates simply isn't true.
Note that in order for this approach to be coherent at all, you need to abandon the idea that it's possible for a voter to vote for a candidate for office. The only kind of vote this system acknowledges is a vote for a political party.