Well, they're two totally different approaches. I'll try to summarize them here, but I highly recommend you look at the NI-VISA Help file in NI-VISA 3.1(or the User Manual in older versions) and look at the sections titled Message-Based Communication and Register-Based Communication. They use different sets of methods.
In message-based programming, as in GPIB, Serial, USB, etc., you send messages (often ASCII, such as *IDN?) down a pipe. You don't worry about exactly where you're sending it - it just goes to the device. You then usually read back a response (again, often ASCII). Binary messages are possible too, but that's usually needed only for a large transfer such as a floating point waveform, where the ASCII representation might be 4-5 times as
long as the binary representation. The device has to interpret each message and convert it to access its registers internally.
With register-based programming, as in VXI or PXI, you have to know the address space (eg, in VXI, A16 or A24 or A32) and you have to know the register offset to write to. Each register address or address range often has unique functionality. It's like writing a device driver and hitting the registers directly. It's lower-level and potentially higher speed, but it's also more bus-specific.
The question of performance really depends where the bottleneck is. Which is faster - the speed of the bus you're using between the PC's CPU and the device, or is it the device's CPU and/or internal bus itself? If the bus is slow, you want to transfer as little data as necessary and have the device do as much as possible, and vice versa.
I hope this makes sense.
Dan Mondrik
National Instruments
