I made some tests. It appears that if the burns are executed even a little adjacent to each other, that is to say, if there is even some seconds' gap between starting times, the buffers will continuously run out and the burning process will be slow.
However, if the burns are executed absolutely simulatenously, then it works just fine with 100% buffer all the time. It appears that if the image file needs to be read from only one point at a time, the number of drives burning the data is not so important. That is to say, there would not be any more bottleneck concerning the hard drive than there usually is with a single burn.
So, that's the trick how Nero does it. I'm sorry to hear that this will not be implemented in Imgburn, since this is such a great software in every other aspect.