I think it's going to be difficult to find experimentally-verified values because the experiments haven't been conducted. (Maybe you want to volunteer :).) There is also a lot of subjectivity here because the accuracy required is going to vary greatly by application and situation, and the amount of APRS traffic in an area is going to dictate how frequently a station can send telemetry without causing congestion on the APRS frequency.
Where I live (southern Saskatchewan, Canada) there are quite few APRS users and only a couple of digipeaters and iGates covering a very large area of land. I have the luxury of being able to send quite a lot of packets without causing flooding. On the other hand, next week I'll be in the Dallas-Fort Worth area and I imagine such frequency of transmission won't be as welcome there because of such a large number of people in such a small area, and presumably, a commensurate number of APRS transmissions plus better infrastructure.
The beaconing on my Yaesu FT-400XDR allows me to configure telemetry transmission by direction change, speed and time. The slower you're going, the fewer packets you need to send. The more often you're changing course, the more often you need to send. At a point, even if you're motionless, you want to send another packet to confirm that you are still in the same location. But what is the exact delay you should use? It's so hard to be concrete because it depends on the purpose. In my case, I send out the packets because I experiment with APRS, I want hams at home to be able to see where I am and I want the opportunity to chat with other hams while driving. Packets sent every five minutes are probably adequate most of the time for what I'm doing. As you point out on your question, though, people on foot doing emergency communications might need to send out packets quite often.
Remember, too - the more packets you send, the more of a hit you'll provide to the APRS transmitter's battery, if its portable. So that's yet another consideration.
So - in summary - I don't think anyone can give a blanket recommendation, just some general things to think about, and then some local experimentation is going to be required.