2. Choosing USB (piano to computer) vs. MIDI (piano to AudioBox) won't affect latency. Latency happens elsewhere, in the audio hardware and software of the computer. So no worries.
This may not always be true. For example, I compared the MIDI latency of my M-Audio Fast Track Ultra, to that of the integrated MIDI interface of a desktop computer. So, for the FTU, the MIDI is of course travelling over the USB to the host computer. In both cases, the MIDI OUT port is looped back to the MIDI IN port. I don't have results yet for pure USB - that is harder to test. However, I do have a means of testing it, however it has some human error. (play notes on a digital piano, and measure the time between finger contact with the key to first audio out from the sound generator. I have used this method in the past with reasonably consistent measurements)
Results, using only note-on/off events:FTU (MIDI over USB)
Message Latency: 7.63ms
Max. Deviation: 11.24msMPU401 (integrated on desktop's motherboard)
Max. Deviation: 0.48ms
From memory, the "message latency" is the average latency, and the "max deviation" is either the absolute maximum latency encountered, OR, the deviation away from the average. (can't remember)
I used this utility:http://earthvegaconnection.com/evc/products/miditest/index.html#download
Now, those results can't possibly be right, because it takes 1ms to transmit a Note-On. I remember having solved that problem - I had to disable "internal timestamps" in the utility in order to get proper results. So, the actual results will be even higher than the above - I will re-test. (I did repeat the tests, but I can't find the results. I pulled these particular results out from a Pianoteq forum post I made)
@Csillag: FWIW, I agree that you probably won't notice any difference, or, perhaps, a slight difference. It probably won't matter which method you use to connect - MIDI through the Audiobox, or directly with USB. However, you may find a slight discrepancy.
If/when I do some further latency testing, I'll publish the results in a seperate thread.