Signatures – Understanding common input ownership heuristics

I understand the common input ownership heuristic is that all inputs in a transaction are assumed to be controlled by the same person. However, given that a signature must be provided for each individual input, multiple parties (each owning an input) can collaborate to create a transaction. This is the basis of CoinJoin.

So why is the common input ownership heuristic relied on so much when performing blockchain analysis and identifying clusters? Besides address reuse (which is much clearer), it is one of the most commonly used heuristics for address concatenation. So why is the common input heuristic discarded once the CoinJoin service is identified, but considered applicable for some random transactions?