A while ago, I worked as a discrete semiconductor component engineer for good old Nortel Networks so I have some knowledge of this stuff, but unfortunately no current access to instrumentation beyond a digital voltmeter.
Good to have someone to talk to about this stuff with then, George :-)
From what you say about their specs it is a mystery. The forward drop is, what, about 2V or so? So that's only 60mW dissipation - not enough to cook them, even totally enclosed. Vibration would be another unlikely cause: if the mounting is soft enough for a filament bulb it's certainly soft enough for an LED.
I wonder if there is high frequency, high voltage spiking from the alternator getting through? Perhaps we should look at something like a TVS or maybe even a simple capacitor to smooth out any such spikes?
My interest in this has spiked a little (forgive the pun!); I'll look into it a little further and see if I can come up with a simple solution. I do have access to a 'scope, so I can look for spikes on my own bike and try out a couple of ideas. I will also talk to my department's electrical technician: he is a wizard and is much more up to date on modern components than I am.