Based on recent experience, I'm not sure it's all that easy to come up with NFRs. It's relatively easy to list the type of NFRs that you want but can be quite hard to get a commitment on a realistic figure for each requirement. You're right that defining a metric for that requirement can then also be difficult!

I'm a bit wary of the "empowered user panel" approach described. Using stakeholders to verify subjectively the system's acceptability seems to risk having a user group with a vested interest in the outcome of the test ("sorry, it's unusable without [pet-feature]"). They will (hopefully) also be familiar with the functional specification and probably not provide a very reliable measure of usability.

I have seen the "expert witness" approach employed to give an opinion on usability. However, the sort of feedback given was more appropriate for iterating the functional specification than signing-off against NFRs. Of course that's not to say that all aspects of usability can be inferred from the functional specification! I think you have to take care to be quite clear in explaining what you're testing for to your expert witness and treat defects, improvements and sign-off criteria from their feedback separately.