Mate,

The tests were run on a moderately fast x86-based system. But I want to emphasize that they are illustrative only; 
actual numbers will be affected by more than just the processor type and speed. The intent of the paper was to 
illustrate the possibilities and that customers should examine the options and decide what is best for them.

The probable reason for the empty start-up case being slower than the 10M database start-up case is that there are no 
databases at all in the empty case, and the empty databases themselves have to be built up from the schema files. The 
assumption is that this takes longer than checking the existing 10M database. In general, the former operation involves 
reading and writing; the latter reading only.

Tim