I think it depends on what kind of job you are looking for. For research universities, I suspect you can get away with it; I actually used a teaching letter based on recitation teaching from graduate school when applying for TT positions 4 years after graduating, because for a variety of reasons, there wasn't anyone else I felt comfortable asking for one from. Honestly, I don't think job search committees put much stock in teaching letters just based on classroom observation. Any value they had has been inflated away by the fact that they are always positive, and to be honest, most mathematicians at research universities don't worry very much about teaching skill when looking at hiring, beyond not wanting to hire someone notably incompetent. If they do, they're much more likely to try to judge from your interview or from student evaluations rather than a teaching letter.
If you're looking into more teaching focused schools, it's harder for me to say. Maybe I'll just leave that hanging and let someone else answer.
EDIT: To address the question of getting two letters, I agree with Noah. I want to be clear that this advice only applies to research universities, but I think getting two teaching letters completely gets wrong the risk/reward calculus for teaching letters. They cannot get you a job, they can only lose one for you. Probably they won't even look at teaching letters until they get to the short list, but if they do, it will be to sort out bad teachers, not to separate competent and good. By far the most important thing about a teaching letter is that it doesn't say anything bad. By getting two, you are doubling this (small) probability for absolutely zero benefit.