Dunno … if I needed to see “an image reflected in a piece of glass,” I would do it in the compositor … really, without thinking further about it. To my general way of thinking, “any single-render’s purpose” is not to hand me, tahh dahhhh, “the finished picture.” No, its purpose is simply to hand me pieces of it, along with the Z-depth and alpha-mask information that I’ll need to use to stitch them all together. Renders are time-consuming, so I really only want to do any of them once and then make the work pay-off as long as possible.
To get the image that will be reflected in the glass, position a camera right where the piece of glass is, looking directly out along the path that the incoming rays of light would take in order to bounce of the glass and into the camera. Put that onto a plane at that location and (just using BI, why not?) shoot one frame. There’s the image you need. A little tweaking of hue, saturation, and intensity, and there you have it. You can fiddle with it, fine-tuning it to be “just right,” and it’s all free-of-charge at that point. I’m no longer rendering, then.
Will it be “optically perfect?” No. Will it be “convincing?” Yes. Good “enough?” Yes. “I see a reflection in it, so it must be glass.”