I think you are right. I have thought a bit about this. I don’t think it makes much sense to add addr:*-Latn because you would need to add it not only to addr:street but also to addr:city and …funny…addr:housenumber
http://osm.org/go/0KoMP0Yrr–
(The letters are really in that order.)
I hope I can explain the reason. First it blow up the database. If you have a street with 100 housenumbers you would store the translation information (addr:street-Latn) 100 times which is already available at the highway-way (highway=, name(:sr)= → name:sr-Latn translation can be made with the street).
The second thing is how I think OSM based routers work. Due to the lack of housenumbers in the most areas in OSM, they look for highway names not addr: names. Chances are much higher to find the highway+name= because there are not so much housenumbers available. When they have the street, they probably look for a housenumber around where addr:street= is equal to name= on a highway=. Second reason is that Serbia would be the only country which has a 2-script addr: tag. Even if someone writes a routing engine trying to support that, then it is not a big deal to translate the addr:*=cyrillic name to a latin name based on the street to which the housenumber belongs… where hopefully name=, name:sr= and name:sr-Latn is added.
There is only reason I can find for adding addr:*-Latn…
http://www.openlinkmap.org/?lat=45.1199562&lon=21.305999&zoom=17&id=927156725&type=node
Such applications parse the address data directly from addr:* tags, so when no addr:*-Latn exist, you would need a workaround.