OK, a few things.

First, about your combined matrix. Are you column or row major order ? In both cases, your matrix multiplication order is wrong. If you're column major, then it should be matBias*matProjLight*matViewLight*matModel. If you're row major, then just reverse it to matModel*matViewLight*matProjLight*matBias. In both cases, the order reads like this: first bring the model into world space (matModel), then into eye/lightview space (matViewLight), then apply the projection (matProjLight), and finally remap the coordinates (matBias).

Second point is the way you handle the w coordinate. In projective texturing, you do not have to do the homogeneous divide by hand, in the vertex shader. Actually, the trick is to do that per-pixel, ie. in the pixel shader. All you have to do within the vertex shader is the matrix multiply, and make sure that all 4 components are given to the pixel processing stage.

In the fixed function pipeline, the perpixel divide is enabled using the D3DTTFF_PROJECTED flag. If you're using a pixel shader, then you have to use the correct form of texture lookup (tex2Dproj, TXP, etc, depends on the shader language you use).