Same thing, bro. I too have this question.. I have asked it in one of my threads:

here i have something that might help you.

http://www.pixelnerve.com/v/2009/02/08/ray-sphere-intersection/

it does ray-sphere intersection that should cover your b).

if you do not want to use opengl, then you need to convert the mouse 2d screen pos to view space. for that you need to reverse the transformations and you'll need matrix data anyway, from OCD, whatever you using.
convert from screen coords to normalized [-1,1]. then transform by the inverse projection matrix. that gives you view space mouse pos that can be plugged to the code above.
if you need it in worldspace, then do the transformation by inverse of modelview (camera matrix).

to cover a) this should also take you there. just use ray-plane intersection tests with a plane on y=0.

hope it helps

i am having the same problem and was wondering if you were able to fix it BlueThen and noncom. the example pixelnerve hints to looks very complex and i would like to know if it is actually worth it to wrap my head around

The solution I used at the time was the code from here:  http://processing.org/discourse/yabb2/YaBB.pl?board=OpenGL;action=display;num=1176483247

It was inconsistent though, and at the current version of Processing it doesn't work at all.

The code you hint to Bluethen seems to work ver nicely for me. It even does more than i had hoped to achieve (recognizing mouseZ on its own). You really helped me a lot with this, thank you very much!
Also thank you for your encouragement noncom :), I hope I will get to a deeper understanding of these perspective projection issues eventually, since I have a feeling this problem will occur to me again from time to time