Could that problem be solved by changing the price at different times, rather than for different people in the same time interval? I wonder if that's why Amazon is making 2.5m price changes a day like the OP says above.
Unfortunately this wouldn't be a proper a/b test since it'd be comparing 1 day of 1 price to another day of 1 price. Only by splitting traffic 50/50 do you properly isolate the price variable.
Most likely, although that solves a different pricing aspect. I would guess most of those are the result of automated pricing optimizations to influence your offers position among other sellers for the same product. There are services that automatically do this for amazon and other price comparison services. It can also result in a book priced at $23,698,655: http://www.michaeleisen.org/blog/?p=358 :-)