What you're describing is the correct behavior. MongoDB stores data
in memory mapped files. So, as you store more data, the files grow
larger, and because they are mapped into mongod's address space, the
size of its process grows larger.
However, the operating system should be paging these out. A
process'es virtual address space can be much larger than the amount of
physical ram on the machine, and pages that aren't currently in use
will be written out to disk. If all you're doing is adding files,
you're going to keep requiring more pages to be loaded so they can be
written to. Eventually, you'll use up as much of the physical ram as
the OS will allow, and it will then start writing out older dirty
pages in order to make room for new ones. This could take some time;
if you are monitoring your system, you'll see a lot of writes as
demands for physical ram from other processes cause this process to
evict its working set. If you're starting up new processes, you can
expect that to take some time as the OS writes out the other pages in
order to make them available for new processes.
Chris