I'm not an expert on the subject, but I've done some testing that could be related to CPU wear a year ago or so.

I ran nbench a multitude of times with several clock/voltage combinations. While running the tests, I monitored current_now as reported by the bq27x00_battery module. Unfortunately I've lost the spreadsheet with the exact results, but I do still remember my observations:

A higher clockspeed with substantially lower voltages required a lot more current from the battery than a lower clockspeed with substantially higher voltages.
Setting a lower or higher voltage for the same frequency didn't nearly affect the used current as much as setting a lower or higher frequency.
This is with SmartReflex turned off in software.

These results seems to support joerg_rw's post that the SoC does have micro regulators that affect the current. This in turn seems to suggests that the user set voltage isn't nearly as much in play as in traditional desktop/laptop overclocking (and thus overclocking the N900 can't be compared to that).

I would be grateful if someone* could reproduce (or disprove) the results of my test. Nbench for the N900 can be found here. Personally, I think the higher current_now is enough 'practical' prove that OC'ing does wear out the CPU much faster than running the cpu at stock speeds, pretty much regardless of whether you're undervolting or not.

* do we have a volunteer?
--

By the way, I don't know how accurate current_now from bq27x00_battery is, but the fact that it went up significantly with higher frequencies at least means something I guess. Testing was done with SSH over USB (offline mode, display off).