Yes, it is true; you can put any old junk into them and with hardware VP you'll most likely still draw correctly. With software you'll possibly crash, or draw triangles-on-LSD, or something else interesting, however. It is important to get them right with software VP as they define a range of the vertex buffer that is used in the draw call (if the params had been named otherwise it would have helped understanding a lot), which suggests that software VP operates in two passes, the first being to run your vertex shader (or other transforms/etc) on the defined range, the second being to take the results of that and transfer it to the GPU - without the ability to define a range the software VP engine doesn't know in advance how much of the vertex buffer is going to be used, so it may pre-emptively transform the full thing.
I definitely second the recommendation to get them right anyway despite this; even if you think right now that you'll never need software VP you may find yourself using it as a fallback at some time in the future.
As an interesting aside, this is exactly analogous to OpenGL's glDrawRangeElements, and suggests that the same behaviour applies there too and that with hardware T&L glDrawRangeElements is no advantage over glDrawElements - which my own informal benchmarking confirms.
Note that these two params were removed from the D3D10+ variant with the rationale being that they're not needed and that most drivers ignored them anyway. There's an MS document floating around somewhere that explains this.