Currently, CHDK just saves the sensor data directly to disk. Although they are named .crw, they aren't actually in the official CRW format.
sRaw appears to just be a reduced resolution image, so you still get the dynamic range of RAW but you get less pixels. Fine if your final product is going to be web or something like that. The advantage would be less with powershots, since they only get 10bpp vs 12 or 14 for the DSLRs, but it still might be worthwhile in some situations.
It would be possible to save a subset of the raw pixels from CHDK, but the CPU and memory system is pretty slow, so proper downsampling might not be any faster than saving the whole file. Saving every Nth pixel (with some gymnastics to account for encoding) should be doable, but would further impact quality. Similarly (and probably a bit easier) you could just save a sub-image (AKA "digital zoom")
Alternately, it might be possible to hijack whatever canon uses to produce the lower resolution jpeg, if that happens before the 10bpp is converted. Where we currently get the raw, it always appears to be full res regardless of what size you selected in the canon menu (on my a540 anyway)