Using CPU utilization, of a switch like a 65xx, with a sup2T, is a rather poor metric. This because, the data plane is handled by dedicated ASICs and the control plane by the CPU. Ideally, you should always see a fairly low CPU utilization, even if the switch is passing lots of data traffic.
Also keep in mind, as you jump up in bandwidth capacities, (at least short term) overall data volumes don't increase, however transfer times should decrease.
You should be able to obtain capacity needs by analysis of your data volume. Then you should analyze if your data transfer times need to be improved. Lastly, you might look at your interface drop statistics.
BTW, although the Nexus series is optimal for large number of servers, the Catalyst series, with the proper hardware, can do quite well too. For example, even your "old" 6500s, if -E models, with a sup2T, offer 80Gbps fabric to line cards. I haven't seen many LAN networks that saturate the sup2T's 2 Tbps fabric.
Your customer might not "need" more performance than their current equipment offers, so similar capacity might be fine, but "newer" is often faster for "equivalent" replacement devices. That shouldn't be assumed, however. Further, newer equipment should be looked at, for its many other aspects (many noted by Leo), too.
