The problem is that the pair programming exercises are not proof of a real work enviroment. Some of the problems or challenges at work won't get an answer in 2 o 3 minutes, sometimes you need to think more of them (in work, maybe one or two days, in the excercise, maybe 30 minutes).
Secondly, some candidates get very nervous/anxious when they have to answer code in an interview and can't show they real knowledge, etc. Pair programming hasn't worked for me because you get a lot of false negatives that are hard to distinguish from the real false.
To your first point - sure, but if you're not willing to invest the one-or-two-days in the exercise why should the candidate? If you're offering a top-tier salary, or something so technically exciting that it can't be resisted, then sure, you get to set the bar as high as they'll jump. Otherwise "Are we willing to invest this much" is a good rule of thumb for what you can reasonably expect of the candidate.
To your second point - any hiring exercise is going to be stressful for the candidate. False negatives are better than false positives (for the hiring party). If you're certain that this is going to cost you too many good candidates, then you're going to have to find some other way to demonstrate good faith up front - offer to pay them, offer to host them on your premises, something like that.
Secondly, some candidates get very nervous/anxious when they have to answer code in an interview and can't show they real knowledge, etc. Pair programming hasn't worked for me because you get a lot of false negatives that are hard to distinguish from the real false.