Well, I guess I'm about to become labeled as a mentally deranged lunatic :) But hey, I am a uni teacher so I have an excuse for that :)
No offense to you or the academics, Peter. But you're one of a few who enable this because you want to show your students. 
Anyway, you probably meant to say that it would be nuts to configure STP PortFast on a trunk, right?
You're exactly right. Enabling portfast on a trunk is so "yesterday", in my opinion. If a trunk port(s) or an etherchannel is configured correctly, there's a significant chance portfast is irrelevant. The speed to get the ports to go from down to passing traffic is really boils down to one or two seconds. Unfortunately, I didn't have the time to "lab" this so I don't have the facts and figures to back my claim up. Happy to see you shoot it down (again, no offense to your or academics).
Well, there have been situations when I felt perfectly fine enabling it, such as trunks to router-on-stick or to a server running several virtual machines. I knew the devices on the other end of the trunk would never perform bridging/switching and would never run STP. In effect, these devices were end devices just like ordinary PCs on access ports. We always do a certain leap of faith when we configure access ports as PortFast ports because under circumstances, a switch or a physical cable loop can be connected there and cause problems. Yet we use the concept of edge ports, and in RSTP/MSTP, we would suffer hard without it. The reluctance of activating PortFast on trunk ports is understandable and I feel a certain kind of hesitation myself but that is caused by the plain fact that trunk links usually go to other switches, and we are aware of the risks. However, if the trunks are provably connected to non-switching devices then they are just as good and eligible for PortFast as access ports.
What you've mentioned above, I can't argue. Unfortunately for me, you've raised a serious question about my philosophy about portfast on a trunk port which I've seen before. Here's the funny thing, though. We've got ESX/VMware stuff. But we've never heard them ask for this feature before. But you know something? I wrote my response in haste. I should've stated "STP Portfast and/or BPDU Guard on an inter-switch link(s) is a bad idea".
I mean it's OK to lab it. But given the complexity of interconnection of different switches to various stuff going around, we're happy with leaving portfast on a trunk port disabled.
PS: Thanks for your post, Peter. Pretty strong (and interesting) points you've raised.
