I've just done some quick testing on this and if the QVD is written with the data sorted by the filtering field, I get a big improvement in load times by using an inner join rather than a where clause, so ...
FROM [DataSorted.qvd] (qvd);
load 424372 as RandNumber autogenerate 1;
... gives a better result than ...
FROM [DataSorted.qvd] (qvd)
If the data is unsorted there is no improvement.