From the trace:

"""
16981 glCreateShader(type = GL_FRAGMENT_SHADER) = 12
16982 glShaderSource(shader = 12, count = 1, string = &"//!!FS
// ======================= Begin fragment shader program ===================================
uniform sampler2D Tex0;
uniform sampler2D Tex1;

vec2 encodeNormal(vec3 n)
{
    float p = sqrt(n.z * 8.0 + 8.0);
    return vec2(n.xy / p + 0.5);
}

vec2 encodeColor(vec3 rgb24)
{
        rgb24 *= 255.0f;
        // nvidia int3 rgb16 = rgb24 / int3(8,4,8);
        ivec3 rgb16 = ivec3(rgb24) / ivec3(8,4,8);
        float greenSplit = rgb16.g / 8.0f;
        vec2 stuff;
        stuff.x = rgb16.r * 8 + floor(greenSplit);
        // nvidia stuff.y = frac(greenSplit) * 256 + rgb16.b; 
        stuff.y = fract(greenSplit) * 256 + rgb16.b;
        stuff /= 255.0f;
        return stuff;
}

void main(void)
{
        vec4 tex0 = texture2D(Tex0,gl_TexCoord[0].st);

    if(tex0.a < 0.5)
    {
                discard;
    }

    gl_FragColor = vec4(encodeColor(tex0.rgb * gl_Color.rgb),encodeNormal(gl_TexCoord[2].xyz));
}

// ======================= End fragment shader program ===================================
", length = NULL)
16983 glCompileShader(shader = 12)
16984 glGetShaderiv(shader = 12, pname = GL_INFO_LOG_LENGTH, params = &580)
16985 glGetShaderInfoLog(shader = 12, bufSize = 581, length = NULL, infoLog = "0:14(11): warning: Float suffixes are invalid in GLSL 1.10
0:17(31): warning: Float suffixes are invalid in GLSL 1.10
0:22(11): warning: Float suffixes are invalid in GLSL 1.10
0:17(21): error: could not implicitly convert operands to arithmetic operator
0:19(32): warning: `greenSplit' used uninitialized
0:19(12): error: could not implicitly convert operands to arithmetic operator
0:21(18): warning: `greenSplit' used uninitialized
0:21(12): error: could not implicitly convert operands to arithmetic operator
0:21(12): error: operands to arithmetic operators must be numeric
")
"""

I tried running it with force_glsl_version=130 -- that seems to help with st/mesa, but it doesn't appear to be hooked up for i965 (??? it used to be! it hought...)

I'm not sure when float <-> int autoconversion was added, maybe in 1.20? But it also uses textureSize, which was only added in 1.30.

BTW, I believe the correct solution here is to add a "force_glsl_version" option to drirc. Should be easy to do with the executable name (hopefully it's not "wine"). Can you test adding something like

        <application name="Hard" executable="Hoard">
            <option name="force_glsl_version" value="130"/>
        </application>

into an appropriate place in your drirc? Probably have to play around with the executable. You can also just stick that into the environment.

Not to ask a stupid question, but...why are you running this in Wine?  There's a native Linux version of the game which renders correctly without any workarounds, at least on i965.

(In reply to Kenneth Graunke from comment #3)
> Not to ask a stupid question, but...why are you running this in Wine? 
> There's a native Linux version of the game which renders correctly without
> any workarounds, at least on i965.

Yes, I know that the native Linux version runs properly on Mesa.
I just thought it was worth noting that this bug still exists under Wine (I observed this bug over a year ago but didn't bother reporting it).