I assume what interests an agent in a project is a combination of the likelihood of selling it to a publisher and the probable number of copies that might be sold. As others have said, there is no typical. If an agent loves the book, thinks she can sell it, she'll take it on. Same for a publisher. If the submission is something that they know is only for a tiny market and there's no money in it (say for example, picture books) they might not take it on. They do have a business to run, but it's not so cut and dry with a formula or anything. A niche book that someone can sell without an agent because there's a small market for it might not be something an agent takes on, but I'd guess that's a small percentage of submissions.
I don't think anyone in publishing buys/takes on a book thinking "Well, it's not bad, but it won't sell a ton of books." They buy what they think will sell well and then see what happens. Some books get a huge advance and are expected to sell millions and flop. Others are bought for pennies and sell millions. No one knows what the market will do, so they just have to buy what they love and think will sell well.
What is the typical percentage of projects that are accepted by agents that end up being published? Again, there's no typical. Some agents have been very successful in certain genres or markets and sell a lot there, but don't do so well in others. But if the agent has been around a while and is with a good agency, then they're selling a decent amount. Otherwise, they couldn't afford to stay in business. But even if an agent has a 100% success record (and I don't think anyone does), that's no guarantee your book will sell, too. There are just too many variables. I'd say, if you write thrillers, query agents who have been successful with thrillers. If you go with an agent who's never rep'd or sold a thriller, your odds are probably lower, but if they're a good agent they can still sell it to a great publisher.
What is the typical cut-off in terms of commission/ number of books sold when it becomes worth the agent's time? There is none. Agents can only speculate what a book will do, same as publishers. They have no clue what the press run will be until they start shopping it to booksellers and orders start coming in. And that happens well after the book is bought. And even if booksellers think a book will do well and order a lot, if no one buys it, it can still flop. Or sell like gangbusters and need multiple printings.
I know it's comforting to have numbers and figures to gauge yourself against, but publishing isn't that kind of business. Although agents and editors have developed their eye to spot good books, there's really no way to tell what a book will do until it's out in the stores and in the public's hands. Books that are "expected" to do well based on what's sold before often get a leg up in terms of marketing and promotion, but again, that's still no guarantee they'll do well if the readers don't buy it.
All we can do is write the best book we can and send it out there.