The biggest reason is that ObjectIDs are 12 bytes, whereas an equivalent string is 24 bytes. Over a large enough collection, those 12 bytes saved per ID really add up! Those IDs also mean fewer bytes transferred over the wire when reading or writing the document, as well.
Additionally, some ODMs expect ObjectIDs for external document references, and may be confused by string versions of the ID. I am not familiar enough with PHP ODMs to say if this might affect you specifically, though.
Regarding the API stuff, though, you should probably be doing normalization of the data before sending it to the client anyhow, because since Mongo doesn't enforce a schema, you can have literally any sort of data in a given field, so you might have some documents that have string IDs, and others that have BSON IDs, and your API would happily send them both through to the client, but one or the other might cause breakage. In this particular case, you should use BSON ObjectIDs in your documents, and then should cast them to strings in your API output.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…