Currently, when we expand the heap in G1 we expand it one region at a time, even if we are going to expand it by a multiple of regions. We might want to optimize that and expand it in a more coarse-grain manner (at least by a multiple of the page size at a time?). I don't think this is a big performance bottleneck. But maybe expanding one region at a time, when large pages are used, might reduce G1's ability to use large pages.