Basically “All of the above” and even more. As user86458 said, it has a lot to do with how the sound travels both inside and outside of the cabinet, which makes both the time- and frequency response, and complex phase response, dependent on the rotation-position and thus on modulation time. What makes it even more complex is that the horn and the woofer rotate separately, and take different times to speed up and slow down. (Though many Leslies have the woofer rotation just completely disabled.)
I found that a Leslie can be modelled quite well as eight sets of impulse-response recordings (two for each microphone and speaker, in orthogonal positions of the movement, respectively). Each of these is like an IR sample you could make of a guitar amp. But then they are mixed together with both time-dependent amplitude and delay, both modulated in sync for each speaker but not in sync between the speakers. All this together creates the rich combination of tremolo, chorus, phasing and resonance/reverberation elements, without however losing the sonically cohesive character of it all working together as one effect instead of separate units for each stacked together. I implemented a Hammond VST instrument whose Leslie works this way – I wouldn't claim it's perfect or as close as commercial Leslie-simulator, not sure what exact tricks these use. But I would say it has unmistakeably Leslie character. That VST can be heard on this recording.
Simpler combinations, like only syncing together the LFOs of a chorus, tremolo and filter or multiple short reverbs with tremolos, do not give results that I found convincing; it really requires that both amplitudes and phase delays are modulated in the complex way that impulse responses provide.