A typical design becomes fraught with peril when trying to use strain guages. They tend to measure the very things that you don't want to measure, like temperature and lateral loads. Whilst purists will scream that you can't allow the chain ring to move with respect to the spider I think that the way to get a useful signal out of a mechanical device is to allow some significant and measurable motion - much larger than the effects of temperature expansion. I'm thinking of a rotation that equates to 1mm/kW at the circumference. Using this relatively large motion we need to measure a movement of about 0.01mm in order to achieve 1% resolution. Using interfering magnetic discs we could have magnetic domain sizes of the order of 0.1mm and easily reach this value. What makes the proposed design insensitive to the usual mechanical tolerances is that the magentic tracks will be balanced - ie there will be two sets of tracks generating interference patterns that move in opposite directions. It is only the relative movement of these patterns that needs to be detected. Also, the interference mixing takes place magnetically before the pickup so there is no need to use high speed, low sensitivity pickups. Instead, simple hall effect devices will detect the two slow beat patterns and measure their relative phase difference. The electronics part of the design is almost trivial. This is a mechanical gizmo...
