Data files, like application binaries, graphics files and so on, are only downloaded if there isn't already a matching version on the client machine. So the fewer variable parts there are for a given type of project, the less persistent downloading...

There's a very large database file which is searched by MCM tasks - that only needs to be downloaded when a new data set is being scanned, so all that usually has to be downloaded for individual tasks is the "Settings File" (which isn't very big at all.)

Some projects (such as MCM) do a very large number of runs against the same data file(s), whilst others may have distinct data files for each batch (MIP1, for instance) so an individual user may only see one or two tasks that use a particular file -- more downloads, but often of much smaller files!

And, finally, some projects may have tasks that depend on the output of previous processing on the same core data (e.g. FAH2) so there are unavoidable single-task data files which might be quite large.

So it might just be that it looks more efficient, but if you were to consider the number of tasks run against a given data file (or files) the amounts of downloading might not be as disparate as they seem to be on a quick glance!

By the way, I am only an end user, so the above is based on observations and exploring the BOINC client_state.xml file and directories...

Happy crunching - Al.

[Edited to add specific reference to FAH2.]

TY. Al and lava. Very interesting and informative