I had considered this free-floating idea also, but the ability to "hamster" while in the sphere rules it out: it tells us gravity works within the sphere as normal (or else the occupant wouldn't be able to hamster around) which means a) the weight of the occupant holds the sphere to the ground and b) if there's no ground underneath, the occupant - and thus the sphere - are going to fall until there is.
Now, what happens if the sphere lands in water?
Lan-"I have this vision of a railway switching yard where the engineers are playing train-soccer with one of these spheres"-efan
You raise a good point, though it is not, strictly speaking, explicit in the spell description. Moving by pushing on the sphere's "walls" [I find the use of this word, in the plural, quite odd.] to roll it might not require a physical interaction with the ground; it might be a further magical effect of the spell.
Even given that gravity maintains its usual hold on the creature, if it, inside the sphere,
does fall after rolling off a cliff, do we know that the sphere and character would accelerate? Or would they continue to move at the half-speed rate of rolling, constrained by the magic of the spell and limited in the kinetic energy that can be imparted to the occupant by the work of gravity?
Then, if they did accelerate, surely the airfoil of the sphere would factor into their rate of acceleration and terminal velocity, right? Using a handy online calculator (
http://www.calctool.org/CALC/eng/aerospace/terminal), I find that the terminal velocity of the sphere, with the occupant's weight in it, would be around 25-30 mph. (I'm making assumptions using a medium-sized creature of 150 pounds). For comparison, via Wikipedia, the terminal velocity for a belly-down person is about 122 mph, depending on altitude and other factors, and for a person with a parachute, it's about 17 mph (about three times the speed of the feather fall spell). According to another handy tool (
https://www.google.com/search?q=mph+to+km/h&ie=utf-8&oe=utf-8) 30 mph (or 48 km/h) would be reached under normal conditions by a person falling approximately 9.1 meters, or just about 30 feet. Thus
if the sphere and content accelerate while falling, and
if they fall far enough to reach terminal velocity, the occupant should take no more damage than he or she would by falling 30 feet under normal conditions.
There is one other unresolved aspect of the scenario, and that is what actually happens when the sphere strikes the ground (or some other surface). I find the hypothesis that the sphere would stop virtually instantaneously and the occupant would take falling damage as normal to be unsatisfactory. We know that the sphere is immune to all damage, and that physical objects cannot pass through the barrier. This means that the force of the impact will be transferred either to the sphere's contents
or to the ground itself. We know that "a creature or object inside can’t be damaged by attacks or effects originating from outside," and the cause (gravity) leading to the effect (acceleration and impact) is outside the sphere. All of the energy of the impact, therefore, should go into the ground, and a small crater should be blown out until the sphere stops--or else the sphere should push the entire planet a minuscule distance--leaving the rider unharmed.