(This is too long to be a comment, too naive to be an answer.)
If $M$ is a Riemannian Manifold, then the group of isometries is a Lie group. This is an old (and apparently difficult) theorem of Myers-Steenrod.
Generally this will not act transitively, if your metric is bumpy or something.
(I think) it's not always possible to put a metric on a manifold for which it becomes a homogeneous space.
Here is a "proof:"
I googled around a bit and found this (a paper by Mostow!): http://www.jstor.org/stable/1969997?seq=1#page_scan_tab_contents
In particular, if a homogeneous space has solvable fundamental group, then this theorem puts a bound on the rank in terms of its dimension. (He defines what he means by rank.)
But now, because (apparently), fundamental groups of 4-manifolds can be pretty much anything you want (they need to be finitely generated), you can probably cook up a counter example.
But you would have to actually read that paper somewhat more carefully than I just did in order to be sure. (I can't read past the first page right now, his definition of homogeneous is on page 3. Maybe it is something completely different!) Let me know!