Parsimony is considered a desirable or even necessary characteristic of a scientific theory but what this means is not clear. There are many types of parsimony (see the article on Simplicity in the Internet Encyclopedia of Philosophy for a list). The most common kind of parsimony is qualitative parsimony, often called Ockham’s Razor, which says that new kinds of entities should not be posited if possible.
Qualitative parsimony privileges atomistic and evolutionary theories since they posit only one kind of atom (with many combinations) and one kind of life (with many variations). This seems arbitrary without further justification. Perhaps Thales has this in mind when he asserted that everything is water in some form. In terms of classification, “lumpers” dominate under qualitative parsimony.
Another kind of parsimony in science is quantitative parsimony, which says that the number of entities should be minimal. By itself this would privilege a theory that posited many classes of entities but few in each class. In short, “splitters” dominate under quantitative parsimony. But quantitative parsimony is rarely mentioned and sometimes even denied.
It would be most fair to accept both types of parsimony, which may be traded off against one another. If a new kind of entity requires many fewer entities, should it not be preferable to positing more entities? And if a small number of additional entities requires fewer kinds of entity, should this not be preferable to positing more kinds of entities?
One way to look at this is that entities, kinds of entities, etc., are explanatory resources that are not unlimited. The best explanation is one that optimally uses explanatory resources. Just as classifications are best that minimize within-class differences and maximize between-class differences, so an optimal approach to explanation is best.