
Of course there's latency & jitter in the midi protocol itself too, just based on the messages being sent, but I accounted for that in my testing. Otherwise I've noted no more than a maximum of 10ms latency on any of these interfaces in the past, and while USB interfaces *used* to be worse than PCI/PCIe based (soundcard) ports I no longer think that's the case with current (core2 or i5/i7) based hardware. The last one is the most affected by system/cpu load as Scope cards process all 'Async' DSP Signals (including MIDI) under a single thread and so heavy Scope processing itself can lag Midi processing.
MIDIMAN MIDISPORT 2X2 DRIVER WINDOWS XP MAC
I used to put effort into time aligning certain things that were layered as closely as possible, and this forms the basis of most of the 'measurements' I took but I did initially sit down about 12-13 years ago and come up with a working methedology (under Logic 4.x & 5.x on both Mac & Windows at that time) for correcting midi jitter. I used several pieces of outboard gear including but not limited to: Roland 909Access Virus KB, Novation Drumstation, Novation Basstation (key), Roland JV1000/JD800/JX3P+PG200/CDX1, Yamaha FS1R, Roland 'VDrums' brain (which was VERY low latency for its internal DSP overhead at the time) and several AKAI samplers.

Midi ports on Creamware/SonicCore Scope DSP cards (under XP/Scope 4.5)

Midisport 4x4 (anniversary & original midiman models) I have tested the following under OSX, WinXP & Win7 64bit: Even with $50 USB midi interfaces I find overall latency to be under 5ms and jitter less than +/-2ms (for a total of 7ms max). There is jitter and latency in many devices, but I stopped trying to measure and correct it years ago.
