What problem do you think that using an interrupt service routine solves?

Also, on behalf of Chris Hobbs:

Apart from the disruption to the OS, one other factor to consider is the time taken to do a meaningful memory test. I 
was working with one customer some years ago who was trying to carry out memory tests dynamically. We calculated that, 
given the amount of memory to be checked, the time required was well over the time for which a memory error could be 
tolerated.

One other problem is that destructive memory tests only find hard memory errors, not soft ones and some experiments 
indicate that soft errors tend to dominate.

Is ECC RAM not an option?

--Elad