Dear Jeremiah and Shovelhead,
To my mind you're wrong. Indeed PhotoScan recognizes the camera used for each picture you import in a single chunk, so it creates 2 different cameras if you import both terrestrial and aerial pictures at the same time. (Even better to enter precalibration parameters if available).
And if you gather all the conditions needed (Overlapping / Computer calculation capacity) to align everything at the same time, DO IT !
Indeed, if you're able to do that, you will be able to balance everything at the same time (calibrations, tie points projections, GCPs coordinates) and your model will become stronger !
If something make this process impossible (I've handled a measurement where my terrestrial set of picture hasn't enough overlap with the aerial survey), then you will have to process each survey separately and then align your chunks together using known GCPs or distinctive point you're able to measure in your main model and then export them into your other one.
But when you align chunks together, only a 7-parameters transformation is available (3 translations, 3 rotations, 1 scale factor) so you won't be able to fix any residual distortions.
About lens, I would say it depend on the kind of survey you handle. I usually use micro-sensors (Phantom 4 Pro UAV, GoPro Hero4) with a reasonable flight height (respectively 30-35m and 3-4m) and I'm able to produce centimeter accuracy models.