Very interesting article indeed, few things I found that might actually hamper the performance considered for the solution in the share market scenario, your concept of implementation creates a scenario where there is one write(the common sproc) for every read, after thinking abt it for some time I thought that may be this can be just queued in some way(through another thread in the front end) and the more critical read executed. Another way it to have decision making channels that essentially dictate the way data is read, eg. stocks in news will be queried more... but then this is probably overshooting, another consideration is how long do the trends persist... you dont want to get into a situation where u r continuously re configuring ur partitions at a time when the trend no longer holds, more like how some people's hair styles are constantly behind the latest fashion 😀

But over all must say it made some interesting reading and the concept still holds.