There's nothing technical preventing using multiple AGP ports with X86 chipsets. It's a combination marketing/bandwidth/driver problem. AGP is just a variation on PCI, and multiple independent PCI busses are commonplace.
For example, a number of Opteron boards have APG, 32bitx33MHz PCI, 64x66 PCI-X, and 64x100/133 PCI-X (four independent PCI busses). And, when they came out a lot of the nVidia drivers failed because they used a fixed bus#/device# (which pointed to the second CPU) instead of polling to find the video card. How well do think those drivers would do with two identical cards on two busses?
Actually, the way that AMD implements their AGP-bridge/hypertransport-tunnel it would be realtively easy to make a board with multiple AGP. And multiple CPU's with integrated memory controllers would help the the memory bandwidth bottlenecks....
It's a shame the market just isn't there.
Maybe PCI-Express will allow some multi-GPU, multi-monitor, ultra-performance solutions.