T3D (1.1b) eating shader constants [fix] and worldToObj bug [also fix] - RESOLVED

I wrote a custom shadow class to replace the projected shadow in my project. This class just re-renders my characters using a custom material that flattens their models onto a pre-defined plane and uses stencil to prevent shadows from blending against each other. It looks good, however my shadows are always centered on their casters, so if someone jumps, the shadows gets off the ground.

I re-wrote my projection over and over again, trying to project in object space, then in world-space, and nothing fixed it. Right now I perform the projection in world space and convert the result to object space before applying the modelview:

#include "common/hlslStructs.h"

struct ConnectV
{
   float4 hpos : POSITION;
};

uniform float4x4 modelview;
uniform float4x4 objTrans;
uniform float4x4 worldToObj;

ConnectV main( VertexIn_P IN )
{
	ConnectV OUT;
	
	float3 planeCenter = float3(0,0,0);
	float3 planeNormal = float3(0,0,1);
	float3 rayVec      = float3(1,1,-1);
	float3 rayOrigin   = mul(objTrans, IN.pos.xyz);
	
	float planeD = dot(planeCenter, planeNormal);
	float cosAngle = dot(rayVec, planeNormal);
	float distToPlane = planeD - dot( rayOrigin, planeNormal );
	float distToProj = distToPlane / cosAngle;	
	
	float3 shadowPos = rayOrigin + (rayVec)*distToProj;
		
	shadowPos = mul( worldToObj, shadowPos );

	OUT.hpos = mul( modelview, float4(shadowPos,1) );

	return OUT;	
}

But the shadows keep moving off the projection plane, so I then decided to look at the shader constants using NV Perf HUD, and then the problem became obvious.

Here's the objTrans shader constant (4x4 matrix at C4) value when the object is drawn normally:
The object is at position (0, 0, 0.11) with scale (0.45, 0.45, 0.45) and no rotation.

Now look at its value when I'm drawing my custom material:
The last row at C7 (the one which should contain the position) is gone! It was overwrriten by the contents of worldToObj. Seems T3D decided that objTrans was a 3x3 matrix for some reason. I'm going to debug the code that sets the constants to see what is going wrong.