VGA cards are definitely *NOT* designed for anything like laser displays.
A VGA card's architecture is designed for fast rasterizing of 3D vector graphics -- the exact opposite of what we're trying to accomplish, which is generating (and transmitting) vector data. So, except for limited purposes (if you use things like GPGPU), it's totally useless for laser show usage.
Especially when you're using OpenGL, which in itself is a perfect way of representing a vector display, but when fed into a video card would only be useful for quick rasterizing (3D rendering). You will actually need the specific glColor(), glVertex() and glBegin() and End() statements to make sense of graphics for laser shows.
High-speed video interfaces like HDMI, Infiniband or SDI are *most definitely* not suitable for laser displays since they're designed to carry rastered pixel data.
Ethernet by itself is a suitable medium to carry the data you want, although IP based protocols (UDP or TCP) might not be very useful. They don't provide the timing guarantee you need to feed your DACs and galvos a continuous stream of position data -- needed for safety reasons. In short, if the link gets congested or disrupted and the flow of data gets interrupted, the galvos will slow down, possibly creating hot beams. A scan-fail will not interfere in such a case since the data fed into the scanners corresponds with the scanner's position -- it will just treat the (unintended) hot beam as an intended one, possibly creating a dangerous situation for the audience.
Without guaranteed timing towards your projectors and their internal DACs, safety is a concern and only something like a PASS board will be able to save you. Unless you supply the projectors with considerable intelligence and (hardware) watch-dog timers which, by themselves, can predict the galvo's motion and beam exposure towards the audience. That also has the advantage of not having to worry about DAC or projector properties (like scan speed, or beam attenuation maps) in your show software -- you just feed vector data into the projectors (through TCP/UDP over IP) and the projectors themselves (which have embedded processors) will handle their specifics and make sure the timings towards the DACs and laser drivers are correct.
Ethernet-based protocols with guaranteed timings do exist, though. Just don't expect it to be based on IP.