Yeah, I pretty much agree with most of your points. Although, I don't think you should read the four points quite as standalone as you appear to have done: it seems more a train of thought to the last one, but in a rough way, not a proof way.
Objects are fairly good at modelling discrete systems. They're not very good at continuous, but then computers in general aren't, and tend to end up treating the problem in a somewhat quantum fashion. So, if you're attempting to replicate a real-world system, you always have some kind of information loss if the system contains any continuous process. And this is all assuming you can capture enough information about the thing to know how it works.
I'm kind of surprised at your thought towards the end though - if anything, I would think it would be more appealing to the analysts, although perhaps your descriptions are slightly different to what I'm thinking. In my experience, OO tends to appeal to people who want to map the software onto the problem directly, essentially making a little world inside the computer. The abstraction side of things is very rarely used in real-world OO, again, IME.