Agreed, speed isn't the primary concern of the challenge. Two reasons I'm looking specifically at this end of things though:
1) Most of the existing teams are now down to blending results, requiring multiple usage of algorithms before the final blend is performed. In the interests of keeping this process fairly sane, storing and retrieving the data efficiently seems logical given it will be accessed repeatedly.
2) Trying to follow what I perceive to be some form of good practice. No point writing code based around one data access method and then discovering I've got to re-write half of it just because I've changed how the data is stored.