Looking at the advertised bandwidth vs bandwidth history from Tor Metrics[1], it appears that guard relays see much more traffic than exit relays. I think it might be partially because guard-only, guard-middle and guard-exits aren't separated, but would it really skew the numbers that much?